Mining and Organizing a Resource of State-changing Verbs

Most knowledge bases (KBs) that emerged in recent years are static. They contain facts about the world yet are seldom updated as the world changes: attributes or possessions of entities in the KB change over time; what (actions) the entities can do or what can be done to them change over time. Current methods for extracting facts in the KBs do not yet detect such state changes although these changes have implications for the correctness of the KB. We propose a method for learning change-of-states in entities that are brought about by event-expressing verbs acting on the entities. These change of states in entities can be viewed as KB updates of facts pertaining to these entities. We envision a resource of state changing verbs, that contains preand post-conditions of each verb on its arguments: the entry condition of entities that allow an event expressed by the verb to take place and the condition of entities that will be true after the event occurs. When a state changing verb acts on an entity or when the entity’s state change is detected in text, we can use the resource to update KB facts about the entity.