The Office of Naval Research is offering this grant to advance Text Analytics for Data to Decisions (D2D).

Today’s warfighter has access to text-based information from a wider range and greater number of sources than ever before. The influx of information can potentially improve warfighters’ situation understanding and decision-making. However, it is clear that there has been, and will be no, increase in the number of warfighters to process and interpret the growing volume of available data. The practical implication of this is that the DoD has access to more data than it can process to achieve actionable information in support of diverse military information needs. It is imperative that we create, harvest and exploit technologies that will help realize the potential for improved decision-making without imposing a need for increased warfighter numbers or their workload. On behalf of the Office of the Secretary of Defense (OSD), Office of the Assistant Secretary of Defense, Research and Engineering (ASD (R&E)), the Office of Naval Research (ONR)) is interested in receiving full proposals for the Data to Decision Program. The program has three (3) primary thrust areas, (1) contextual understanding, 2) event prediction, and 3) machine translation and processing.

Together these thrusts areas seek to develop new technological capabilities that support military operations . The three (3) thrust areas are described below:
Thrust Area #1- Contextual Understanding
The 2012 National Security Strategy has indicated that “for the foreseeable future, the United States will continue to take an active approach to countering [threats] by monitoring the activities of non-state threats worldwide.” With ever increasing decentralization of decision-making brought about by ubiquitous electronic communication, there is a need to identify threats in complex, uncertain, contradictory and incomplete large data sets available in the open source environment. Irregular warfare, non-state terrorism movements, and uncertain environmental patterns that trigger major weather disasters are examples of events that require military response. In responding, decision makers will use text analytics to develop the necessary contextual understanding of the region and key elements. Strategies for achieving contextual understanding can include observational data, a priori knowledge models, and inductive knowledge. Contextual understanding is generally achieved through a combination of human and computer processing techniques that take advantage of a person’s cognitive ability to fuse and assimilate multiple sources and types of information for new insights. Correlation and aggregation of open source data, such as agriculture, weather, terrain, demographics, economics, social patterns etc. is nontrivial but vital to effective military response.

This research thrust area seeks innovative approaches to the following aspects of contextual information 1) the discovery of specific events that are planned or may have occurred, 2) stated values and beliefs that motivate behaviors of interest, 3) discovery of topics and concepts developed in a shared community, 4) analysis of semantic relationships existing in a community, 5) strength of relationships, 6) community structure and clusters of social networks, and 7) semantic analysis and trending of emotional support expressed toward topics or persons.

Thrust Area #2- Event Prediction
Intelligence analysts need the ability to rapidly monitor and analyze event information in large volumes of unstructured textual data, such as news articles or Human Intelligence (HUMINT) reports, in order to achieve and maintain persistent Situational Awareness (SA). For example, the ability to stay apprised of events that have already occurred, as well as events that is threatened or planned, would be a valuable contribution to an analyst’s SA; however, the amount of unstructured textual data available is well beyond what can be manually read and processed in the time available. A capability enabling analysts to rapidly extract event information from large volumes of unstructured text and store it in a structured form, such as a database, is needed to improve an analysts’ ability to maintain persistent SA.

The goal of this thrust is to advance the state-of-the art for extracting events with their attributes of modality, polarity, genericity, and tense from large volumes of unstructured text. Modality of an event indicates if the event was a real occurrence. Examples of event modality include asserted, i.e. “The bomb exploded on Sunday;” believed, i.e. “It is rumored he will be sentenced;” hypothetical, i.e. “If he were arrested, he would be convicted of murder;” and threatened, i.e. “He threatened to attack the country.” Event polarity indicates whether the event actually occurred. For example, “The city was not attacked” is an event with negative polarity, and “The attack occurred on Sunday” is an event with positive polarity. Genericity indicates whether an event is specific, i.e. “The city was attacked on Saturday”, or generic, i.e. “They specialize in transporting weapons.” Tense indicates whether an event occurred in the past, is occurring in the present, or will occur in the future. Secondary challenges include, but may not be limited to, rapid customization to different sources/styles/formats of textual data, and rapid customization to various domains. While addressing other technology gaps that would contribute to the capability would be useful, it is of lower priority to the program since it should not happen at the expense of addressing the primary research challenge of extracting event attributes.

DOD text analytics will often be focused on social groups who have an interest in hiding behavior, such as terrorist networks. In such conditions, innovative methods are needed to identify proxy features of a network that may aid discovery goals to uncover potential events of interest. Temporal trends are one such category. Examples may include factors such as frequency of contacts between nodes or clusters, inter-contact time, recurrent contacts, time order of contacts along a path, and delay path of information diffusion. Methods to extract, characterize, and monitor social networks dynamically over time is a research challenge of interest that may support event prediction.
Scalability and predictability have been perennial problems in certain types of text-based analysis, such as social network analysis (SNA). As networks increase exponentially in size and complexity, it is harder to use graphical methods to represent, monitor, and understand network behavior. The representational graphs grow to unmanageable size, contain complex relationships among nodes, and often contain several varieties of nodes. Two promising approaches are being explored by the SNA community, visual analytics and semantic analysis. Visual analytic methods supported by ontology have been shown to reduce the visual complexity of these graphs to enable users to identify important structural and semantic aspects of networks. Research is needed to identify key actors and supported relationships, detect the presence of bridging nodes that can uncover hidden sub-networks, and determine the flow of resources (information, money, influence) within the social network.

This research thrust seeks innovative approaches to the following aspects of event prediction 1) identify proxy features of a network, 2) extract temporal trends, i.e. frequency of contacts between nodes or clusters, inter-contact time, recurrent contacts, time order of contacts along a path, and delay path of information diffusion, 3) extract, characterize, and monitor social networks dynamically over time, 4) evolve visual analytics and semantic analysis at scale, 5) identify key actors and supported relationships, and 6) detect the presence of bridging nodes that can uncover hidden sub-networks, and determine the flow of resources (information, money, influence) within the social network.

Thrust Area #3- Machine translation and processing
Unfortunately, while there have been notable advances in information retrieval (e.g., intelligent, adaptive and ontology-based search engines), data mining, and (to a limited extent) development of cognitive aids and decision support tools, progress on the “ingestion” or understanding of information has not advanced as rapidly as for collection. This is due, in no small part, to the huge increase in unstructured information available to the warfighter. While text can be extremely valuable, it is not readily amenable to automated processing. Instead text must, in large part, be handled, assessed and interpreted individually by humans. As a result, the sheer magnitude of these resources can overwhelm the very mission they are collected to support.

Many areas of the world where future military action may be required are rich in language or dialect diversity. To fully engage local populations and respond to humanitarian needs, language translation will become critical to text analytics efforts. Strategies that lead to computationally efficent algorithms are needed to develop and improve technologies for machine translation and processing, information extraction, and automated summarization. Also relevant, are the methods and algorithms to develop and improve technologies that “import” physical sources into electronic form such as optical character recognition (OCR) and speech recognition as input to machine translation and processing, information extraction, and automated summarization. Development of language data in support of building these technologies and development of metrics to evaluate underlying software algorithms are also needed.

Research in the areas of linguistics, natural language processing, mathematics, statistics, computational data analysis and visualization, computational sciences and computer science are of interest. In addition to the application of research methods and approaches, it is important to evaluate the impact of these efforts areas with regards to the way they change how data is collected, analyzed and assessed to meet a prescribed time for operational necessity and efficiency. It is of value to use open standards to reduce costs. This research thrust seeks innovative approaches to the following aspects of machine translation and processing 1) intelligent, adaptive and ontology-based search engines, 2) improved data mining, 3) improved cognitive aids and decision support tools, 4) “ingestion” or understanding of information at scale, 5) improved information extraction, 6) improved automated summarization.

**************************************************************************************************The FULL ANNOUNCEMENT is available on the Grants.gov website by scrolling to the top of the synopsis page and clicking on the “FULL ANNOUNCEMENT” box surrounded by the dotted line at the top of the page.