【作者】 车万翔； 刘挺； 秦兵； 李生；

【Author】 Che Wanxiang Liu Ting Qin Bing Li ShengInformation Retrieval Group, Harbin Institute of Technology 15001

【机构】 哈尔滨工业大学信息检索组；

【摘要】 本文通过对经分词和词性标注的大规模语料库(1.8GB)的统计,计算出语料库中出现的词对个数、距离及方差,并应用t检验的改进方法,得到了词对之间的“搭配强度系数”值R,以此来衡量它们之间这种搭配关系的强弱.这一系数直接面向依存文法分析,以此得到一个句子中各个词的搭配关系强弱序列表,以后将要从此表中得到依存文法树.目前我们可以在智能搜索引擎等多种场合找到此种方法的应用.更多还原

【Abstract】 In this paper, we statistic a very large corpus (1.8GB) and work out the word-pairs’ number, distance’s mean and variance. And then we use a modified t test method to fetch a "Collocations Coefficient" R in order to weigh how strong of their relationship. This coefficient orients to the analysis of dependency grammar straight. In this way, we have a word-pairs list in a sentence sorted by R’s order. Later, we can get a dependency grammar tree from this list. Now we can find several applications using this method such as intelligent searching engine, etc.更多还原