Michael Gasser, Eliana Colunga and Linda B. Smith
Relations lie at the center of humankind's most intellectual endeavors -- science, mathematics, poetry. Relations are also fundamental to any account of linguistic semantics. Indeed, a fundamental psychological distinction between objects and relations may be reflected in the universal linguistic distinction between nouns and verbs [GentnerGentner1982,LangackerLangacker1987]. Despite the importance of relations in understanding cognition and language and despite the dogged and continuing work of psychologists, linguists, and philosophers on the problem, there is no well-accepted account of the origins of relations. What are relations made of? How are they made? This chapter addresses this question by considering the psychological evidence on how children acquire relations and relational language and by proposing a computational model.
The plan of the chapter is as follows. First, we consider past proposals of how relations are represented and the implications of these representational ideas for development. Second, we review the developmental evidence in the context of five psychological facts about relations that must be explained by any account of their origin. This evidence suggests that relational concepts are similarity based, influenced by specific developmental history, and influenced by language. Third, we summarize Gasser and Colunga's Playpen model of the learning of relations. This connectionist model instantiates a new proposal about the stuff out of which relations are made and the experiences that make them. Finally, we outline how the model explains the five psychological facts and consider the implications of this model for the question addressed in this volume, the interface between conceptual structure and spatial representation.