Date: Wed, 20 Nov 1996 19:16:58 GMT Server: Apache/1.0.3 Content-type: text/html Content-length: 1696 Last-modified: Sun, 03 Mar 1996 15:56:27 GMT
How do children learn natural language, and how could one get a machine to do this? One popular view is that the task is so daunting that the child (and the computer) must come with considerable knowledge "wired in". But it is not clear how this wiring could be implemented in a developing nervous system or even what the appropriate knowledge would be. An alternative is an approach which starts with a particular architecture and then acquires the necessary knowledge in response to input from and interaction with the world. This project explores the acquisition of words within this second sort of framework, focusing on their form: how words are composed out of constituent morphemes (morphology) and how the primitive sounds of a language combine with one another (phonology). The goal is a general neural network architecture with the capacity to learn words in any language. This topic is tied to a number of general issues in cognitive science: the origin of modularity, the nature of temporal short-term memory, the relation between perception and action, the development of apparently hierarchical knowledge.
Associated Faculty: Michael Gasser
For more information click here