Language as a source of evidence for theories of spatial representation.

David Marr's metatheory emphasized the importance of what he called the computational level of description--an analysis of the task the visual system performs. In the present article I argue that this task should be conceived of not just as object recognition but as spatial understanding, and that the mental representations responsible for spatial understanding are not exclusively visual in nature. In particular, a theory of the visual system must interact with a theory of the language faculty to explain how we talk about what we see--and how we see all the things we talk about as though they are part of the perceived world. An examination of spatial language both raises the bar for theories of vision and provides important hints for how spatial understanding is structured.