Computational Support for Concept Blending applied to Musical Instruments

Abstract

This thesis presents a concept blending implementation that suggests which properties of a known concept are most compatible to blend with another concept. The implementation uses Wikipedia descriptions of concepts as data source, and NLP tools such as WordNet and Stanford CoreNLP to create a representation of concepts and their properties. By joining the generalized properties of two concepts in a tree structure, we look for patterns in the data which correlate with properties that make sense to blend. A heuristic function based on these patterns is used to rank the subtrees to return top suggestions of features to blend between the two concepts.