High Level Fuzzy Labels for Vague Concepts

Vague or imprecise concepts are fundamental to natural language. Human beings are constantly using imprecise language to communicate each other. We usually say ‘John is tall and strong’ but not ‘John is exactly 1.85 meters in height and he can lift 100kg weights’. Humans have a remarkable capability to perform a wide variety of physical and mental tasks without any measurements. This capability partitionsof objects into granules, with a granule being a clump of objects drawn together by indistinguishability, similarity, proximity or function [8]. We will focus on developing an understanding of how we can use vague concepts to convey information and meaning as part of a general strategy for practical reasoning and decision making. We may notice that labels are used in natural language to describe what we see, hear and feel. Such labels may have different degrees of vagueness. For example, when we say Mary is young and she is female, the label young is more vague than the label female because people may have more widely different opinions on being young than being female. For a particular concept, there could be more than one label that is appropriate for describing this concept, and some labels could be more appropriate than others. A random set framework, Label Semantics, was proposed to interpret these facts [3]. In such a framework, linguistic expressions or labels such as small, medium and large are used for modelling. These labels are usually defined by overlapping fuzzy sets which are used to cover the universes of continuous variables. Different from Computing with Words [9], fuzzy labels are usually predefined and used for building intelligent systems such as decision tree [4, 5], naive Bayes learning [7] and rule induction systems [6] without involving the computing of semantic meanings of these labels. In this paper, we extended the label semantics framework with high level fuzzy labels. In previous research of label semantics, fuzzy labels are used