Tell Me About

Our proposed work addresses the significant challenges in providing responses to complex"tell me about"questions that intelligently consolidate relevant information into a finished dossier. We will induce the semantic signature of the question target by using relevant sample intelligence reports to derive a set of information requirements which will be used to create a finished dossier. These information requirements will drive the retrieval of relevant textual passages which will be semantically enriched using novel approaches to relation discovery and frame induction. The entities and events discovered in these enriched passages will then be linked to a structured knowledge base allowing for the addition of pre-processed factoids about related entities and events. Finally, this information will be fused into a final, finished dossier which will provide a rich source of links and information for the analyst. We will explore a hybrid strategy with ties to research in Automatic Wikipedia generation that will augment and enhance LCC"s existing state-of-the-art Ferret QA system. In addition, we will leverage existing state-of-the-art natural language understanding and content extraction in order to acquire the semantic and pragmatic information necessary to satisfies the demands of today"s information analysts with a single finished dossier.