Natural Language Text Retrieval Using A Large Semantic Network

In the 1980s, two independent developments in text retrieval attempted to improve Boolean searching. These were statisticaly searching, which allowed users to enter requests in plain English and retrieve documents based upon the statistical occurence of words; and concept based searching, which requires users to define concepts in a knowledge base and perform searches by entering a concept by its name. Both of these methods made some improvements in accuracy, but placed additional burden on the user. Statistical searching has not been readily accepted because it is difficult to explain the results. Concept based searching requires extensive labor to build the underlying "knowledge base" of search concepts. ConQuest Software, by using retrieval software, called ConQuest (TM), that includes the best of both statistical and concept based searching, but also eliminated the need for users to build a knowledge base. The objective of this effort is to combine ConQuest with new processing to reach breakthrough levels of precision and recall accuracy by exploiting the power of a large semantic network that includes, but is not limited to the contents of WordNet. The research results will demonstrate that linguistically based text retrieval systems have a higher upside potential than purely statistically based systems. ANTICIPATED BENEFITS: The proposed system is useful to the Government in many ways: it enhances research programs such as TIPSTER, it provides a capability for intelligence agency work, it provides document management, and in manages information (DTIC). The commercial market for text retrieval is rapidly growing and has applications spanning corporate intelligence, legal, regulatory, marketing, and document management.