Can a semantic network improve information extraction? This paper outlines LaSIE Information Extraction (IE) system's approach to knowledge representation. Unlike many IE systems that use domain-specific patterns, LaSIE translates sentences to a quasi-logical form and constructs a discourse model to derive template fills. At the core of the system is a ‘world model’, represented as a semantic net, which is extended during the text processing by adding the classes and instances described in the text. The system's knowledge representation formalisms are described, their use in the IE task, and how the knowledge represented in them is acquired. Preliminary evaluations of the approach indicate performance comparable to shallower methods. The paper explores the system's knowledge representation formalisms, their use in the IE task, and how the knowledge represented in them is acquired, including experiments to extend the system's coverage using WordNet. This generality and extensibility offer a route towards the higher precision required of IE systems to be genuinely usable technologies.
This paper aligns with Natural Language Engineering's focus on computational linguistics and natural language processing. The use of semantic networks for information extraction is a key topic in the field.