Structuring information from natural language descriptions: Accounting for uncertainty

In this paper, we report on a part of a large experiment in structuring information from natural language descriptions of animals from a children's dictionary. The structuring included the recognition and postulation of properties and capturing the is-a hierarchy from the descriptions. The material was taken from the 2007 edition of the American Heritage First Dictionary. We applied the methodology and resources of the Ontological Semantic Technology, developed for language and information processing computational systems to represent the meaning of text. We then proceeded to account for uncertainty in order to improve the crisp results. We have demonstrated it on the example of the descriptions for the feline family of animals.

I. Introduction

There is a very tempting hypothesis about uncertainty: since uncertainty occurs when the information is not detailed enough, then refining the conceptual grain size should reduce the uncertainty. The hypothesis can be illustrated, for instance, by this exchange of cues because the answer leaves it uncertain for the questioner what exactly the other person is doing: