The topic of this paper is the conceptual organization of polysemous prototypical qualities. This study, based on data collected in 24 languages, makes use of a single notional space composed of 110 notions. This space enables us to separately represent the polysemies observed in each language as well as polysemous patterns observable in several languages in order to contrast the variability specific to each language with the linguistic invariance. The results show that what is common in the language sample is based on recurring polysemies organized in networks. This method will also be useful in explaining how the linguistic variability is built up. Indeed, some of the qualities involved in these networks always take part in polysemous associations specific to only one language. Such qualities, called federative notions, are characterized by the fact that they are regularly involved in polysemous patterns, and across numerous languages.