Event Representation in Knowledge Systems with Context Hierarchies
Scone is a high-performance, open-source knowledge-base system intended for use as a component in many different software applications. The Scone architecture supports contexts, which allows efficient representation and reasoning about different states of the knowledge base. Events and actions are particular types of knowledge involving mulitple world-states. The thesis is intended to present a flexible and powerful event representation that may be used in natural language processing, plan reasoning and generation. The use of contexts in representing events will allow efficient reasoning, while the use of background knowledge will allow better verb understanding in natural language.
The general event representation will have two associated contexts representing the before and after states of the world (or the preconditions and effects of the event). Events, like other entities in Scone, will be a part of a multiple inheritance network, allowing a specific type of event to be instances of many types. The virtual copy semantics of Scone, allows fast and efficient inheritance and reasoning of the before and after contexts.
Compound events, or plans, involve a more complex interaction between the contexts of events. A sequential plan requires certain contexts to be merged, and certain roles to be unified. Again, Scone allows efficient ways of representing compound events. The thesis will explore the representation of plans and possible uses of it with different software applications.