Extracting conceptual knowledge from text using explicit relation markers

This paper describes a method for extracting knowledge from large corpora using conceptual relations such as definition and exemplification, The two major steps in this process are the identification of specific relations using positive and negative triggering, and the extraction of the conceptual information by combinatorial pattern-matching. Validation of extracted candidate text is performed by analysis of part-of-speech tag patterns. The algorithms are embodied in a robust program which is capable of attempting extraction even in the absence of part-of-speech tags id the input text. Unlike many knowledge extraction systems, the KEP program is designed to be non domain specific, Intended applications described include knowledge acquisition for automatic examination question setting and marking, and knowledge acquisition for the creation and updating of semantic nets used in a hypermedia-based tutoring system.