Distributed Server Architecture - We will develop a distributed server architecture for ontology construction and use based on ontology servers which provide access to the contents of ontologies via a network API and to information derived from the contents by a general purpose reasoner. Ontology servers will be analogous to data base servers and will enable distributed ontology repositories and distributed servers for editing, browsing, etc. which access the repositories. Ontology servers will provide a suite of services, including configuration management for ontologies, support for ontologies that have components resident on remote servers, and support for an Ontology-URL that enables ontologies to be linked into World Wide Web pages so that they are accessible for browsing like any other Web "document".
A particularly difficult task for an ontology server is supporting efficient query answering from ontologies represented in a highly expressive language. To provide that support, we will develop an idiom-based retrieval facility that returns instances of a sentence containing schema variables from a given ontology. The retrieval facility will employ a general purpose reasoner that can be run as a background process to infer and cache sentences that match idioms used by the API and by translators. These derived facts will be removed by a truth maintenance facility when the statements on which they are based are removed.
Representing Essential Knowledge In And About Ontologies - There are significant gaps in the expressive power of current knowledge representation languages. These gaps prevent the inclusion in ontologies of knowledge about domains that is essential for many high-priority applications and knowledge about ontologies themselves that is essential for effective ontology use and reuse. We will close some of the more important of those gaps by developing new representation formalisms, integrating existing formalisms, and incorporating the results into the tools and servers developed in the project. The results will enable ontologies to contain richly textured descriptions that include uncertainty, are structured into multiple views and abstractions, and are expressed in a generic representation formalism optimized for reuse. In addition, a computer interpretable ontology description language will enable annotation of ontologies with assumptions made, approximations made, topics covered, example uses, competency, relationships to other ontologies, etc.
One of the fundamental gaps in the expressive power of standard knowledge representation paradigms is their inability to represent and reason with uncertain information. Uncertainty is unavoidable in battlefield and crisis management situations, where sensors and other information sources are invariably unreliable, and where the dynamics of the situation are always unpredictable. We will design an uncertainty representation language which allows the incorporation of uncertain information into a knowledge base. The focus of the work will be on augmenting frame-based languages which support a concise representation of the properties of objects and classes of objects. We will provide mechanisms for representing uncertainty about slot values, and the probabilistic dependencies between the values of different slots within a frame. We will also support the use of probabilities in component hierarchies, coherently relating the probabilities over the properties of an object with the probabilities over the properties of its components. Similarly, we will support the use of probabilities in inheritance hierarchies, providing mechanisms for inheritance of probabilistic properties from a class to its subclasses.
This uncertainty representation language will support the modeling of uncertainty in complex planning problems. By representing an action schema as a frame, and its preconditions and effects as slots, we can model the uncertainty about the behavior of an action within our language. Furthermore, component hierarchies can then be used to represent the composition of a high-level action from low-level actions, while inheritance hierarchies can model alternative decompositions of a high-level action. By allowing the composition of low-level actions in these two ways, our system will support reasoning about complex plans.
We will also develop a representation for ontology competency based on evaluable functions and relations analogous to methods in object-oriented programming. We will develop tools for composing and associating with an ontology methods that are procedural implementations of theorems provable in the ontology. The methods would then be directly callable by applications that use the ontology.
Ontology Construction - We will address key difficulties in building large scale ontologies by developing tools for specifying the overall structure of an ontology during the early stages of development, supporting teams of collaborating developers, testing and debugging ontologies, merging ontologies, and automatically acquiring probabilistic domain models from data. The focus of the work on probabilistic domain models will be on learning probabilistic parameters, thereby alleviating the difficulty of acquiring such parameters from human experts. However, we will also investigate the possibility of learning the qualitative dependency structure from data.
We will develop tools for testing ontologies that enable a developer to use an ontology to describe familiar situations and to query those situations to determine if the situations as described have expected properties. The query answering facility will use the ontology server's general purpose reasoner to derive answers.
We will also develop tools for merging ontologies that describe a common sub-domain using differing vocabularies, assumptions, approximations, views, abstractions, etc. The tools will use the ontology server's reasoner to derive and add to the merged ontology equivalence, subsumption, and disjointness relationships among the classes, predicates, and functions of the ontologies being merged; will provide facilities for renaming classes, predicates, and functions in the merged ontology; and will provide facilities for combining classes, predicates, and functions that have differing definitions but are intended to be equivalent.
Obtaining Domain Models From Large-Scale Ontology Repositories - Sophisticated retrieval, extraction, composition, and translation tools will be needed in order to effectively obtain domain models from large-scale ontology repositories that satisfy a set of application-specific requirements regarding content, level of abstraction, view, underlying assumptions, representation language, useability by problem solving methods, etc.
Given that a set of classes, objects, relations, functions, views, and/or topics have been identified for retrieval, relevance-based extraction and composition techniques are needed for producing an ontology which contains all the sections of repository ontologies that are relevant to the identified elements. We will develop such techniques by extending our research on irrelevance reasoning in knowledge based systems and compositional modeling in engineering domains.
For ontologies from a repository to be incorporated into an application system, the knowledge must be translatable in some practical way into the receiving system's representation language. Currently, knowledge base translators are difficult to build, maintain, and extend because they must be hand coded by experts and translation rules are typically embedded procedurally in the program. We will address these problems by developing vocabulary translation tools that enable a knowledge base builder to specify and apply declarative translation rules, a suite of translators for the extended representation language we will develop, and a declarative translation rule language that enables customization of the translators for particular uses.
We will also develop tools and techniques for retrieval and extraction of base-level probabilistic domain models from a knowledge base expressed in the uncertainty representation language. These models will be tailored to the situation at hand, allowing the system to appropriately focus on the relevant aspects. The models constructed will be hierarchically structured Bayesian networks, a natural augmentation of standard Bayesian networks. Thus, the system will support the use of a knowledge base by standard Bayesian problem solvers, as well as by more specialized problem solvers that can take advantage of the additional hierarchical domain structure.