© 1994 by Oxford University Press
Articles |
The Learning of the Past Tense of Danish Verbs: Language Learning in Neural Networks
University of Aarhus
Roskilde University Center
This paper reports the results of simulating the learning of the past tense of frequent Danish verbs on a relatively simple neural network. The purpose was to test the ability of different networks to generalize the acquired knowledge to unfamiliar verbs. The main hypothesis is that this ability is due not only to simple, statistical frequency, but is also rooted in the system's internal representation of the verbs' phonological features. In order to test this hypothesis, different ways of training and testing the networks were used. The conclusion is that neural networks are able to generalize the past tense form from the base form, and that phonological form plays a significant role in generalizing. This questions the scope and validity of the models of learning that are currently adhered to in both first and second language learning theories. The generalization of the results to human language learning, though, is constrained by the artificial learning and testing conditions.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. Hulstijn Towards a unified account of the representation, processing and acquisition of second language knowledge Second Language Research, July 1, 2002; 18(3): 193 - 223. [Abstract] [PDF]
|
|