RC3 Conceptual spaces - a geometric framework for representing concepts


The framework of conceptual spaces [1] as proposed by Peter Gärdenfors provides
geometric means for representing concepts:
A conceptual space is spanned by a set of quality dimensions (e.g., temperature, time, hue,
pitch) and points in this space represent observations made by an agent. Dimensions can be
grouped into domains, i.e., sets of dimensions that inherently belong together. For instance,
the dimensions hue, saturation, and value can be grouped together into the color domain.
Concepts and their properties can then be represented by regions in this high-dimensional
The framework of conceptual spaces has applications both in linguistics (where the domain
structure is used to derive word classes like adjectives) and in artificial intelligence (where it
can be used as a way to link symbolic and subsymbolic representations and processes).
After having explained the overall framework of conceptual spaces, I will introduce some
of its applications. This includes my own research which is focused on concept formation,
specifically on the following question: “How can an artificial agent learn about meaningful
regions in the conceptual space purely from perceptual data?”


cf. above


Gärdenfors, P. 2000. Conceptual spaces: The geometry of thought. MIT Press

Course location

Lecture Room 2

Course requirements


Instructor information.

Instructor's name

Lucas Bechberger




From 2009 to 2016 I studied Computer Science at the Karlsruhe Institute of Technology (KIT)
where I graduated with distinction in April 2016. Since June 2016, I am working as a research
assistant at the Institute of Cognitive Science at the University of Osnabrück. Since October
2016, I am also enrolled in the Cognitive Science PhD program in Osnabrück. In my research,
I try to enrich the conceptual spaces framework by a mechanism for concept formation, i.e.,
for automatically creating concepts from observations.