Proper name knowledge acquisition for text understanding

Current work in proper name analysis is focused on identification and limited categorisation of names. Some research has been carried out in acquiring knowledge of proper names from the contextual information within texts. In this study, we investigate how to transform human-oriented compilations, which contain a rich knowledge of proper names, into formallyrepresented knowledge for computer consumption. If a dictionary is considered the source of knowledge about common nouns, an encyclopaedia should be regarded as the source of knowledge about proper nouns. Considering the amount of work on knowledge acquisition from dictionaries, comparatively little work has been carried out on extracting knowledge about proper nouns from encyclopaediae. Here we discuss the knowledge with which proper names are related and present our methods for knowledge acquisition from dictionaries of biography. Our analysis of biographical entries leads to the observation that there are indeed repetitive patterns used in biographies, in both English and Chinese. These patterns provide the possibility of transforming the knowledge of proper names from text descriptions to encoded knowledge. Taking a sublanguage approach, we report how the content of a dictionary of biography can be mapped onto a knowledge base with a minimum amount of human intervention. ABKAS, Archetype of a Biographical Knowledge Acquisition System, has been implemented. It comprises the pre-processing unit, the sublanguage parser and the knowledge base constructor. The key component of ABKAS is its sublanguage parser which consists of a number of finite state machines. Based on the local grammar we have identified, the sublanguage parser achieves the syntax-semantic mapping. Biographical entries in both English and Chinese are parsed and their corresponding logical forms are generated and further represented in knowledge bases.