Building a Terminological Database from Heterogeneous Definitional Sources

An obstacle to understanding results across heterogeneous databases is the ability to determine conceptual connections between differing terminologies. In this paper, we present the two step approach which we have used to build a terminological database in order to address this issue. First we automatically built a heterogeneous collection of terms and definitions from two types of dynamic sources: 1) glossaries automatically identified from 147 government web sites and 2) definitions extracted from 600 unstructured articles. After storing terms and their definitions, we semantically analyzed the definitions to store the terminological knowledge in a relational database. Currently the database contains 12,780 definitions of 8,431 terms. 1 Motivation for a Terminological Database Due in part to the rapid growth of the Internet, individuals and researchers have unprecedented access to data from many sources. Failure to properly understand the concepts behind these data sets may lead to erroneous conclusions in data analysis. We have built an automated system that can discover and make available concepts, definitions and inter-definitional relationships from several Internet sources by building a terminological database. In building this database, we reflected two desiderata for terminological resources: 1) to enable representation of the ongoing evolution of language since new words and senses continually appear in language, and 2) to allow users to explore the richness of terminological knowledge (e.g how terms relate to each other, what are their semantic properties and attributes). Our goals are to provide easy and efficient access for users of multiple information sources across government sites. Towards meeting these desiderata, as a first step, we automatically built a heterogeneous collection of terms and definitions from dynamic sources. As a second step , we semantically analyzed these definitions in order to identify relations among terms and their attributes. We then built a relational database. The architecture of the system is shown in Figure 1. Terminological database ParseGloss