As we have just seen, languages look very different from one another with respect to space. We would like to know whether the differences really matter for the acquisition process. Views such as the symbolic position of Pinker and Jackendoff would predict little effect: since all children start out with the same universal spatial categories, they should go through roughly the same stages in acquiring spatial language. In particular where a language does not encode one of the universal categories in a direct way, we would expect over-generalization errors in which a particular form is applied to the universal category. Work by Bowerman and colleagues [Choi and Bowerman, 1992] on the acquisition of English and other languages has shown that this is not the case. Korean children, for example, use no global semantic categories of CONTAINMENT and SURFACE CONTACT/SUPPORT, categories which are not expressed in Korean in a transparent way. Instead they learn the Korean distinction between TIGHT and LOOSE FIT early on. The data seem to support the view that the particular semantics of the lexicon of the target language has a significant effect on the way the language is learned.
We think this view is correct. Thus any model of the acquisition of spatial language must account not only for (1) the developmental path babies follow in learning spatial terms, but also for (2) the interaction between the particular lexicon of the language being learned and the way it is learned. This is an argument for a model of the type shown in Figure 3, one in which linguistic meaning and concepts are not clearly distinguished.