Reasoning by Analogy Using Holographic Conceptual Projection

摘要

This thesis discusses the designing of an architecture which mimics a human thought mechanism. The architecture is called a Holographic Conceptual Projection, which uses analogy and dynamic pattern matching combined with some natural language understanding. Our main hypothesis is that we project our way of thinking into words and sentences which we manipulate when thinking verbally. This means we can exploit the structure of sentences to build an algorithm that models our thought mechanism. In our Holographic Conceptual Projection Architecture we give examples of every word within the context patterns. The patterns contain sentences that describe the 'condition', 'desired situation', 'proposition' and 'outcome' of the concept. The concept's patterns are then compared with new cases to see analogies. This comparison is done with dynamic generalization and specialization techniques. Finally after building an implementation, we tested it on an intelligent file-management system and an image processing application.