An Action Ontology Framework for Natural Language Interfaces to Agent Systems An Action Ontology Framework for Natural Language Interfaces to Agent Systems

We present in this article an ontological framework, which is specifically aimed at the development of natural language interfaces to agent systems in task-oriented applications. The ontology framework focuses on the representation of action concepts, since they provide, on the one hand, a suitable basis for natural language processing through the use of verb-centered case frames as semantic representation, and, on the other hand, they allow a connection to the action repertoire of the artificial agent system, which is supposed to respond to a user’s verbal requests with suitable actions. Action concepts are arranged in a taxonomical hierarchy and described in a formalism related to description logic. They are characterized through a set of parameters, which refer to object concepts in the knowledge base, and additional, formula-valued attributes specifying preconditions and effects, which define the dynamic aspects of an action 1 An Action Ontology Framework for Natural Language Interfaces to Agent Systems concept. This integrated representation of action and object concepts in the ontology allows us to provide a well-founded and grounded semantics for action concepts. The ontological framework has been used in the development of a variety of virtual agent systems with natural language interfaces, including an interior design system, an automated cook, a navigational and informational guide for our Computer Science Department, and several simulated household robots.