Memory-based planning knowledge schema acquisition from natural language explanation

Abstract

A natural language understanding system requires extensive knowledge about the world. Since most systems simply have the built-in knowledge schemata, it is appropriate to consider how to acquire these knowledge schemata automatically. In this thesis, the problem of acquiring planning knowledge schema from natural language explanation is addressed by constructing a computer model which analyzes narratives containing planning knowledge. The system attempts to construct the causal structure of the narrative in terms of the goals and their plans. It acquires the planning knowledge schema if there exists. This approach is not the inductive but one trial learning. The acquired schemata contain processing knowledge which is useful in dealing with the kinds of planning, and serve as episodic memory structure indexing scheme.