In Playpen language is crucially involved in the learning of spatial concepts. Spatial concepts are learned as the network picks up on correlations between spatial features of pairs of objects in the world. When spatial terms are presented, these take part in the correlations as well. Because spatial terms in different languages correlate differently with spatial features, the particular language being learned will affect the way spatial concepts are learned in the network. Particular correlations in the world may or may not agree with the words being learned, and this can result in two sorts of effects. Words that match the correlations in the world are learned faster than those that do not. When words agree with the values on a particular visual/spatial dimension, the network can use the language to help it process visual scenes, even in ostensibly non-linguistic tasks.