Closed terminologies and temporal reasoning in description logic for concept and plan recognition

Description logics are knowledge representation formalisms in the tradition of frames and semantic networks, but with an emphasis on formal semantics. A terminology contains descriptions of concepts, such as scUNIVERSITY, which are automatically classified in a taxonomy via subsumption inferences. Individuals such as scCOLUMBIA are described in terms of those concepts. This thesis enhances the scope and utility of description logics by exploiting new completeness assumptions during problem solving and by extending the expressiveness of descriptions. First, we introduce a predictive concept recognition methodology based on a new closed terminology assumption (CTA). The terminology is dynamically partitioned by modalities (necessary, optional, and impossible) with respect to individuals as they are specified. In our interactive configuration application, a user incrementally specifies an individual computer system and its components in collaboration with a configuration engine. Choices can be made in any order and at any level of abstraction. We distinguish between abstract and concrete concepts to formally define when an individual's description may be considered finished. We also exploit CTA, together with the terminology's subsumption-based organization, to efficiently track the types of systems and components consistent with current choices, infer additional constraints on current choices, and appropriately restrict future choices. Thus, we can help focus the efforts of both user and configuration engine. This work is implemented in the scK-REP system. Second, we show that a new class of complex descriptions can be formed via constraint networks over standard descriptions. For example, we model plans as constraint networks whose nodes represent actions. Arcs represent qualitative and metric temporal constraints, plus co-reference constraints, between actions. By combining terminological reasoning with constraint satisfaction techniques, subsumption is extended to constraint networks, allowing automatic classification of a plan library. This work is implemented in the scT-REX system, which integrates and builds upon an existing description logic system (scK-REP or scCLASSIC) and temporal reasoner (scMATS). Finally, we combine the preceding, orthogonal results to conduct predictive recognition of constraint network concepts. As an example, this synthesis enables a new approach to deductive plan recognition, illustrated with travel plans. This work is also realized in scT-REX.