Using categorical grammars and a non-model-theoretic semantics to build automated representations of concepts: A non-keyterm approach to information retrieval.

Abstract

This research develops an information retrieval system (IRS) using a semantic document representation derived using a combination of categorial grammars and conceptual semantics. Of particular interest to this research are documents in the document collection which a user would include in the set of retrieved documents, if the set was selected manually, yet are excluded by automated methods of IR. This research vigorously embraces the belief that language is a comprehensive system which encodes and transmits information. Therefore, traditional keyterm approaches used as a automatic document representation generation paradigm are rejected. Rather, natural language processing (NLP) and computational linguistic techniques are explored in depth as methods to automatically extract information from natural language texts and build electronic document representations. Important aspects of this project include implementation of an automated syntactic categorial parser and a conceptual semantic model to create a working IRS. Results of this research have proved promising and suggest that richer, more complex document representations can be applied to the automated information retrieval (IR) process.