(please see the main page for schedule information)
We are developing a framework for conceptual retrieval (named FACIT) that uses a generative lexicon approach for knowledge extraction. Techniques for knowledge extraction, which attempt to (semi-)automatically populate complex schemas (e.g., cases, lessons, plans) from text documents, can accelerate the development of intelligent decision support systems. In contrast to traditional information extraction approaches that use shallow natural language processing techniques and simple lexicons, knowledge extraction requires a combination of deep natural language techniques and rich lexical resources to support robust and accurate interpretation of specialized subject matter text. Popular lexical resources such as WordNet are representationally inadequate and largely incomplete for knowledge extraction. Consequently, we are investigating a novel alternative approach, involving Generative Sublanguage Ontologies (GSOs). GSOs extend generative lexicon theory, and provide a rich representation for selecting unanticipated meanings of terms and their novel combinations by applying a small set of sense generation operators. I will introduce a case-based application of FACIT and describe our GSO concept representation, which includes novel entity and event structures within an object-oriented inheritance framework. I will also discuss how the GSO representation of world knowledge should improve compound noun interpretation and resolve prepositional attachments. This concept representation enables morphological semantics to significantly reduce the number of entries in a GSO and improve text interpretation.
This work is being performed in conjunction with Technical Lead Dr. Kalyan Moy Gupta, who works with me as an ITT Industries contractor.
Dr. Aha leads NRL's IDA Group, whose focus is on the research and development of state-of-the-art decision aiding tools that can be transitioned to their sponsors' organizations. Their current projects concern knowledge extraction from text documents, intelligent lessons learned systems, hypothesis elaboration for suspected terrorist activities, a testbed for evaluating learning-embedded decision systems on (e.g., gaming) simulation tasks, and plan de-confliction and air vehicle management. David has published frequently on these and related topics, given keynote presentations at several international technical conferences, (co-)organized 14 international conferences/workshops, has served on the editorial boards for JAIR, Machine Learning, and Applied Intelligence, (co-)edited two special journal issues, and serves on several AI conference program committees.