Extracting Lexico-semantic Relations from Document Titles Using Sublanguage Analysis

The sublanguage analysis methodology is based on the theory that texts generated by a group of people possess their own lexical, syntactic, and semantic characteristics. The main thrust of the research reported here is to apply the methodology to Korean document titles, reflecting the fact that there has been little research in applying the methodology for the Korean language although much research has been done with English texts. The research had the goal of finding a way to construct the lexicosemantic representation from the document titles and laying out a firm ground for automating the entire process by developing necessary components and analyzing the linguistic phenomena. We conducted intellectual analyses and extracted a number of lexico-semantic relations between concepts existing in document titles. For each relation, we generated relation-revealing patterns (RRPs). A program was written to extract relations by using RRPs as well as concepts connected by a relation. We applied a chart-parsing algorithm for the problem of determining the boundary of concepts. In our experiments with journal titles, we achieved about 80.9% of accuracy in identifying the relations and the boundary of concepts connected by a relation.