Subsumption and recognition of heterogeneous constraint networks

Terminological knowledge representation (TKR) systems, such as KL-ONE, are widely used in AI to construct concept taxonomies based on subsumption inferences. However, current TKR systems are unable to represent temporal patterns or recognize instances of such patterns from ongoing observations. Motivated by applications such as service personnel dispatching, and plan recognition for interactive user interfaces, we extend TKR by introducing terminological QME (qualitative, metric and equality) networks. In QME networks, nodes are TKR concepts and arcs are qualitative constraints between temporal intervals associated with nodes, metric constraints between end-points of temporal intervals, and equality constraints among roles of different concepts. We use QME networks to represent patterns, and define QME network subsumption, which enables us to organize a pattern library into a taxonomy. We also develop a terminological approach to predictive pattern recognition based on subsumption and a related notion of compatibility. We assign a modality of “necessary”, “optional” or “impossible” to every pattern as events and constraints are observed. We also show how to augment a pattern library for complete recognition. This work, implemented in the T-REX system, enables more sophisticated applications of TKR technology