How do words get their meanings? Does this process depend on the details of what the words are about, that is, on the way things actually are in the world? If it does, does it also depend on the mechanisms within learners which allow them to deal with and understand the world, that is, with their sensory/perceptual and motor systems? And if this is the case, does the way the world is understood by learners depend in turn on word meaning and how it is learned?
These are questions that have been around for a long time. They bear on issues as fundamental as what language is and what symbolic cognition in general is. In this paper we will argue briefly that much can be gained by assuming that the answer to all three questions is yes: regularities in the world and the mechanisms of perception and action matter for language, and language in turn matters for cognition. Then we will discuss some methodological consequences of taking this position, and we will consider what this position means for a particular semantic domain, that of spatial relations. Finally, we will present Playpen, an evolving connectionist model designed to simulate the development of spatial cognition and spatial language. We believe that the fundamental questions can only be answered by such a model, one which makes concrete predictions about the behavior that children actually exhibit.