Understanding and structuring NL descriptions: The case of 101 animals

Our premise is that an intelligent system should be able to structure all the information that can be obtained from natural language text, and it should do it in such a manner that the structured information be useful for further processing. This paper presents an experiment in structuring information from the natural language incomplete descriptions of 101 animals collected from a children's dictionary. The goal of this experiment is to use computational semantic analysis of the natural language descriptions-which are quite heavy in common sense knowledge-and to come up with a hierarchy of and similarities among the described animals, as well as flagging descriptions that are largely inconsistent with other information.

I. Introduction

On the premise that an intelligent system should be able to structure all the information that can be obtained from natural language text, and do it in such a manner that the obtained information be useful for further processing, this paper presents an experiment in structuring information from the natural language descriptions of 101 animals collected from a children's dictionary. We assume that reasoning involved in a child's analyzing such a natural language description is an important part of human cognition, and that children do use their cognitive abilities to generate a sufficiently complete structure from these descriptions, combined with their prior outside knowledge.