English Generation from a Knowledge Base and Overly General Classes.

In this paper, we describe the range of sentences that we synthesize from a knowledge base (KB) using a natural language generation (NLG) system. We give an overview of our NLG system, and then focus on the specific challenge in handling overly general classes such as Tangible-Entity and Physical-Object. Such classes are unavoidable in the KB, and it is undesirable to show them in an output for end users. Our approach has been implemented in an intelligent textbook application that helps students learn better.