Recognizing Event Attributes in Unstructured Text (REACT)

In Phase I of REACT, we will demonstrate how extraction of attributes dealing with modality, polarity, and genericity can enhance the quality of information provided by a state-of-the-art event extraction and coreference system to provide actionable intelligence to analysts. We will demonstrate how our research makes significant improvements to the understanding of polarity and veridicality, the characterization of events as generic or episodic, the inference of author perspectives, and the fusion of event attributes across mentions to enhance the knowledge surrounding and improve the analyst"s situational awareness. We plan to leverage existing state-of-the-art natural language understanding and content extraction capabilities including (1) wide coverage event recognition and coreference resolution, (2) open-domain, customizable information extraction, and (3) methods for determining the social intentions of authors in text. This will enable us to extract events and the rich forms of semantic and pragmatic information expressed in their attributes in order to find and fuse information that satisfies the demands of today"s information analysts.