These simulations were motivated by the goal of explaining two facts well-documented in the literature on children's early word learning: (1) the fact that nouns labeling concrete objects are learned faster than the dimensional adjectives that label the perceptible properties of those same objects and (2) the fact during the protracted course of learning dimensional adjectives, children seem to recognize that the dimensional adjectives comprise a class in that they confuse adjective meanings but do not confuse noun and adjective meanings.
The principal contribution of the present results is that they show that these two facts can emerge from the simple effects of similarity-based learning and thus that they do not demand an explanation in terms of prior conceptual knowledge of noun meanings or the differences between nouns and adjectives. The argument for pre-linguistic notions of the distinction between objects and their properties is often couched in terms of arguments that ``one cannot get something from nothing'' (see, for example, Markman, 1989). These simulations demonstrate that one can get a lot from ordinary effects of similarity and redundancy on learning --- a noun advantage and proto-syntactic categories that in terms of their outward manifestations, that is, performance, look very much like the developmental trajectories of children learning common nouns and dimensional adjectives.
In the remainder of this paper, we discuss the further contributions and limitations of the research by addressing three questions: (1) Are the real nouns and adjectives that children learn like the idealized nouns and adjectives presented to the network? (2) Does the network instantiate a conceptual bias for noun-like meanings? and (3) Could the simple associative effects between linguistic inputs and linguistic outputs be the basis for a more conceptual understanding of the differences between nouns and adjectives?