Event Attribute Recognition and Labeling (EARL)

Today"s intelligence analysts are overwhelmed with textual data in the form of Human Intelligence and open source data from the web. The amount of texts that analysts have access to is far greater than one could ever read. This is a fundamental problem for analysts who have pressing strategic and tactical deadlines. Innovations in event extraction help resolve this problem by turning unstructured text in to structured data stores of events. However, with millions of events in a database, simple event extraction does not sufficiently contribute to analysts"Situational Awareness (SA). To truly increase SA, events must be searchable based on how, when, and if they occurred. This requires the ability to automatically recognize event attributes with very high accuracy. Under the Event Attribute Recognition and Labeling (EARL) effort numerous innovations towards high quality event attribute extraction are made. The EARL approach will use a multi-task classifier that simultaneously labels all attributes, with better accuracy than existing approaches. The EARL approach exploits deep linguistic features extracted from unstructured text. Finally, the EARL approach radically augments the data available, inexpensively and efficiently, by using crowdsourcing. The result is a capability that will far exceeded the current state-of-the-art in event attribute recognition.