Software agent technology is currently based on fixed ontologies and languages, hand-crafted for a particular application. The advent of massively distributed systems however calls for not only a common language between all agents involved but also the ability to autonomously adapt and form concepts about novel experiences and events. We propose a method by which agents can autonomously form new concepts grounded in their own experience. This is an improvement on previous approaches because it can tackle a much wider range of conceptual types and provides an efficient, accurate representation which can be used in a rule based system or conceptual framework.