| A Connectionist Account of the Object-Substance distinction in Early Noun Learning Young children learning vastly different languages generalize nouns in the same systematic way. One task that has demonstrated this is the Novel Noun Generalization task. In this task, children are shown an exemplar, it is named, and then the children are asked what other things are called by the same name. Young children extend the name of a solid object to other objects that match the exemplar in shape irrespective of their color, size or material. However, when generalizing the name of a non-solid substance, children consider material to be more important. In brief, young children seem to know that objects and substances are fundamentally different and classified in different ways. Where does this knowledge come from? One proposal is that it comes from innate ideas reflecting an ontological difference between objects of substances. In this research I present evidence for an opposite idea: that it is learned. I propose that the concepts of object and substance emerge as the natural product of associative learning and generalization by similarity. More specifically, I propose that the formative associative mix consists of correlated perceptual properties, syntactic cues, and most notably, the lexical category structure of early learned nouns. I use connectionist networks to simulate both children's performance in the Novel Noun Extension task and the knowledge they bring to the task. In nine experiments with children, neural substance are created by the statistical regularities inherent in the early noun lexicon by: (1) demonstrating that a connectionist network trained kinds, (2) demonstrating that these associative processes can explain previous findings in the literature, and (3) generating, testing and verifying new predictions. |
