Ontologies and Probabilities: Working Together for Effective Multi-INT Fusion
Kathryn Blackmond Laskey, Terry L. Janssen · 2007
Formal ontologies are becoming an essential tool for Intelligence analysis. An ontology provides upper- and domain-level category systems for decomposing and relating objects, object attributes/properties, temporal events, and relations of interest to the intelligence analyst. A great deal of intelligence analysis focuses on understanding and reacting to instance-level report data of varied fidelity on numerous kinds of entities, events and relations. Some of these entities and relations may not be represented explicitly within the ontology’s categorical structure, but may be ingested from ancillary systems with which the ontology must interoperate. To an increasing degree, intelligence analysts rely on fusing reports from many different sources. These include reports from different kinds of sensors, processed intelligence from various systems and databases, and human intelligence. The common vocabulary and precisely specified semantics of formal ontologies is a critical enabling factor for interoperability. The promise of multi-INT fusion is that individually noisy and unreliable indicators can be brought together to form a common operating picture (COP) of a given situation [1]. Because the reports being combined may vary greatly in quality, it is essential to account for source quality in combining reports. This requires understanding data quality and applying methodologies for combining information that make use of data quality in a sound and principled manner. Probabilistic reasoning is a wellunderstood, theoretically sound, and generally applicable method for combining evidence from multiple sources of varying reliability. Computational probabilistic reasoning is a wellestablished and growing field of research and application (e.g., [2, 3, 4]. Probability has shown its value across a wide range of applications, and many qualitative and heuristic approaches to combining information have been explained as “fast and frugal” approximations to the normative probabilistic solution [5]. Until recently, there has been little research on marrying the fields of formal ontology and probabilistic reasoning. However, this situation is changing (e.g., [6]). This paper will address the question of how formal ontologies can best be combined with probability theory to provide theoretically sound and practically useful semantic technology for multi-INT fusion. We will investigate theoretical concerns associated with the connections between logics associated with formal ontology (e.g., description logic, common logic, first-order logic) and those of probabilistic mathematics. The goal is to provide a high-level discussion of the issues involved with combining ontologies and probabilistic systems as a basis for dialog between these two communities, and to identify a broadly construed research agenda for their mutual development and interaction. The authors of this paper argue the necessity of articulating a clear theoretical foundation as a basis for later development of specific methodologies and languages. An important question, therefore, is how probabilistic formalisms such as Bayesian Networks can be merged with formal ontologies. Probabilistic theories produce qualified conclusions, graded by numerical measures of plausibility. By contrast, formal ontologies have focused on purely logical reasoning that leads to definite conclusions. Formal ontological categories are