Textual Inference for Grounding Events in Space (TIGRESS)

ABSTRACT: In Phase II of TIGRESS, we will continue development of a prototype event geolocating system, known as TIGRESS, which leverages an ensemble of open-domain text processing and sophisticated spatial reasoning capabilities in order to associate events with any and all spatial information that is extracted or inferred from disparate collections of unstructured text. We plan to extend the state-of-the-art spatial event reasoning system developed in Phase I of this effort with a combination of novel techniques, including supervised, semi-supervised, and unsupervised algorithms, which will increase the amount of domain knowledge about events and event structures that informs the event geolocating system. We expect that by integrating forms of spatial inference with robust statistical and symbolic models of the implicit structure of complex events, we can achieve a level of performance for event geolocating which goes beyond what is possible with the more"shallow"approaches which focus solely on the spatial properties of individual events and event structures. In addition to pursuing unsupervised approaches to knowledge acquisition from text, we also plan to explore a tighter integration between the TIGRESS prototype and LCC"s family of fully-customizable information extraction systems (including CiceroCustom and CiceroEvent), as well as knowledge base tools (Lorify). This will incorporate the rich semantic models created by users of these interactive systems into an integrated spatial reasoning framework. BENEFIT: We believe there is a growing need for systems capable of providing a wide range of accurate spatial and temporal information about events from unstructured texts within multiple sectors of the U.S. government, including the Departments of Defense, Department of Homeland Security, Federal Bureau of Investigation, and the national intelligence organizations overseen by the Office of the Director of National Intelligence. We anticipate that the prototype location-tagging and spatial reasoning system being developed as part of TIGRESS will not only enhance the quality of Language Computer"s CiceroCustom suite of open-domain customizable event extraction tools, but will also serve the location-tagging and spatial visualization needs of operational customers as well. Transitioning our technology into these organizations will support intelligence analysts, incident managers, intelligence preparation for the battlefield, and event detection / response collaboration. To accomplish our objectives, we will solicit our current contacts within the Intelligence and Homeland Security communities to demonstrate and tailor this capability to their needs. For DoD customers, we will work through our strategic partners to demonstrate and integrate spatial visualization into their current and future systems. As location-tagging and spatial reasoning capabilities become more robust, we expect the quality and coverage of our applications will drive demand for LCC"s products in the civilian commercial sector as well.