Automatically Constructing Verb Class and Extraction Pattern from Untagged Corpus

摘要

Information Extraction (IE) is the task of identifying, collecting and normalizaing information from NL text. Many IE systems use a pattern-matching approach, but the patterns have to be created for each target task or target domain. In order to improve IE systems and make them portable to various domains, we should have a mechanism to create pattern easily. We will propose a method to automatically construct verb class and pattern from untagged corpus. First, we give to the system seed verb class. Then the system selects sentences appearing verb class. Second, it makes a pattern based on selected sentences and then selects setences based on made pattern. Then it adds on verb class pattern appearing sentence.