Inference and Learning in Computer Model of the Development of Language Comprehension in a Young Child
Mallory Selfridge University of Connecticut
The question of how children learn to understand and speak language has long been acknowledged as one of the central problems to those interested in the nature of language. Recently, computers have provided new tools to those interested in child language, enabling them to embody their theories in computer programs. This chapter reports on one such program, developed to model one child, Joshua, as he learned to understand simple commands between the ages of 1 and 2 years. It is a summary of work presented in Selfridge ( 1980).
This program is based on the notion that children bring to language learning a substantial repertoire of previously acquired concepts and world knowledge, and also well-developed learning and inference abilities. It further rests on the notion that children hear language in situations that enable them to infer its meaning, and that they learn language through such situations. The program described here is given the kind of world knowledge that Joshua had at about age 1, and is equipped with inference and learning rules that appear plausible for that age. It is exposed to the same kind of experiences with language that Joshua had, and learns to understand simple commands comprised of action, object, and relation words to a degree comparable with Joshua's abilities at age 2.
This program was based on observations of a single child. I spent an hour to an hour and a half with Joshua per week during his second year, and tape recorded my observations of his behavior. The decision to study only one child in detail, rather than a number of children in less detail, was based on the belief that the case-study approach was most appropriate for a computer model. In order to provide a computer with a conceptual base for language learning of the sort