Semantic Parsing and role Labeling In Combination Effort (SPLICE)

To Counter Weapons of Mass Destruction (CWMD), DTRA must analyze data from numerous sources about a diverse set of technologies and activities. The extreme diversity in content and the volume of DTRAs data combine to produce one of the most challenging analysis problems in the government. The CWMD mission therefore requires deep analysis - an in-depth understanding of all available data and correlation of information across all sources This analysis challenge can be met by Deep NLP - methods of language processing that develop representations of whole-text meaning and connect those representations of meaning to external knowledge. The goal of this effort is to operationalize the Deep NLP capabilities examined in Phase I of this effort and deliver two Deep NLP-enabled capabilities to DTRA. The first capability is an information retrieval and knowledge-base population tool that provides high-accuracy and high-recall identification of relevant data. The second is a complex even extraction capability that accumulates information about events described over long segments of text spread across multiple documents, and populates event models based on the deep understanding of the way events unfold.