Vision and communication with embodied cognitive systems Geert-Jan M. Kruijff Language Technology Lab German Research Center for Artificial Intelligence (DFKI) Saarbrucken, Germany gj@dfki.de People are embodied -- they have a physical experience of the context they are situated in. This embodied experience is often reflected in the interaction between people, situated in the same context: they can talk about the things they feel, see, and hear. Dourish (2001) calls this the tangible dimension of embodied interaction. Crucial for human interaction is that we can translate what we experience into qualitative statements, which we can talk about with others. One of the big challenges in creating communicative, embodied cognitive systems is precisely this qualitative step. Artificial systems perceive the world in analogue fashion, not in terms of qualitative and usually relatively vague predicates. In this talk, I discuss several cases in bridging the gap between vision and language, and show how we can address these cases in an embodied cognitive system we are building in the EU FP6 IP "Cognitive Systems for Cognitive Assistants". Cases include incremental learning of structural description of objects, and spatial relations between objects in a visual scene.