The goal of this chapter is to introduce distributed representations of language especially multimodal distributed representations that are trained by jointly processing language and perceptual data among which visual data are the most prominent when dealing with spatial language. Multimodal representations have the capability to capture physical knowledge of sizes, locations, distances, etc. in the meaning representations. A necessary condition for this to happen is grounding language in physical reality. This chapter is largely devoted to the concept of grounding, where we distinguish between self-grounding, cross-modal grounding, and situated grounding. A second part of this chapter focuses on what actually has been realized with regard to multimodal representation learning and their abilities to take part in the spatial reasoning tasks discussed in Chap.  4 .