In this extended abstract we discuss four related characteristics of semantic spaces as the standard model of meaning representation in quantitative semantics. We argue that these characteristics are challenged from the point of view of social web communities and the possibilities which they offer in terms of exploring semantic \emph{and} pragmatic data. More specifically, we plead for a reconstruction of the weak contextual hypothesis in order to account for non-linguistic, pragmatic aspects of context. Finally, we mention two consequences of such a pragmatic turn, that is, in the area of named entity recognition and of language evolution.