CISE Research Resources: Purpose-Driven Natural Language Processing

The College Of New Jersey, Ewing NJ

Investigators

Abstract

EIA-0130798 Miroslav Martinovic The College of New Jersey CISE Research Resources: Purpose-Driven Natural Language Processing The unifying theme of the three projects (WHAT, QASTIIR, and Hopewell) is the integration of methodologies of computational linguistics with statistical techniques and their cooperative application to question answering and information retrieval systems. QASTIIR (Question Answering System Through Intelligent Information Retrieval) considers where and when techniques of computational linguistics could best improve performance of a hybrid statistical/linguistic question answering system. WHAT (Web Host Access Tool) addresses client-side personalization of web search queries, exploiting semantics to disambiguate keyword meaning. The Hopewell project automates K-12 curriculum mapping (essentially a digital library of curriculum resources achieved through teacher consensus). NLP/IR techniques should reduce the need for middle layer expertise to manipulate map databases. All three projects address how to disambiguate logical and set expression queries. QASTIIR takes the most theoretical approach, developing theory through established protocols for testing. The WHAT domain provides a highly ambiguous real-world domain using well established databases. Hopewell project provides a second real-world domain whose semantics are more constrained, but that requires a database design responsive to novice querying. Anticipated outcomes are (1) new theory development in natural language processing, (2) highly private user profile development, (3) application of NLP technology to digital library development and retrieval.