A hallmark of human cognition is our capacity to talk about what we see- especially, the objects, motions and events around us. How is this accomplished? Given that language and spatial representations are likely to be quite different in kind, the challenge is to understand how such apparently different systems of knowledge map onto each other, and how these mappings are learned. One solution is to seek core communalities in the structure of linguistic and non-linguistic representations, which could afford learners and mature users a means for talking about their spatial experience. I will discuss this possibility with respect to the structure of paths within events. Over a broad range of event types, formal linguistic analyses, computational models, and experimental studies show that a fundamental property of path representation-an asymmetry between source and goal paths-is pervasive in our non-linguistic understanding of events as well as our encoding of these in language. This suggests that there are deep structural homomorphisms between spatial language and non-linguistic spatial representation. Such structural parallels may provide a partial solution to the problem of mapping dissimilar domains onto each other, and may even yield insight into how some aspects of language evolved.