I will consider several hypotheses about the relationship between spatial cognition and language as they might interact in the development and mature use of spatial terminology. In particular, I will lay out some hypotheses that suggest different degrees of interactivity, and will begin to evaluate them in the context of recent evidence from individuals with Williams Syndrome (WS). This genetically-based syndrome presents a highly unusual cognitive profile, combining profound spatial deficits with relatively spared language. This raises questions about how these individuals learn to talk about space. If the spatial representations normally thought to underlie spatial language are disrupted or impaired, can the learner nevertheless acquire part or all of the semantics of spatial terms, and if so, are the implicated mechanisms different from those underlying normal development? The Modularity Hypothesis suggests that linguistic and non-linguistic representations of space are modular and separate and do not interact in the development and mature use of spatial language. The Partial Homomorphism Hypothesis suggests interactivity between the two. Different kinds of evidence suportt each hypothesis and these differences suggest several challenges. First, we should consider the possibility that different semantic properties of spatial language may emerge in development through different learning mechanisms-- some requiring grounding in non-linguistic spatial knowledge and others not. Further, we need to refine our ideas about what kinds of spatial knowledge might, in principle, give rise to space-language interaction effects. Given that spatial cognition is not one global system, it is likely that properties redundant across systems may be the ones specially engaged by language. These considerations have implications for understanding the nature of the language-space interface.