Modeling language constructs with compatibility intervals

Abstract

We describe a representation for modeling meaning of natural language constructs that is closely related to fuzzy sets. Same as fuzzy sets, it allows to express quantitative relationships between different concepts, and is designed to support vagueness and imprecision common to natural language. We compare the representations using several examples, and argue that in some cases the proposed representation may be a good alternative to the fuzzy set based representation, and that it may also be easier to learn from data.