MACHINE LEARNING AND THE COGNITIVE BASIS OF NATURAL LANGUAGE
Shalom Lappin
King's College London
Friday, April 15, 3:30PM MJH Rm 126
Sponsored by the Stanford Humanities Center/Mellon Foundation
Graduate Research Program
The past fifteen years have seen a massive expansion in the application of information theoretic and machine learning methods
to natural language processing. This work has yielded impressive results in accuracy and coverage for engineering systems
addressing a wide variety of tasks in areas like speech recognition, morphological analysis, parsing, semantic interpretation,
and dialogue management. It is worth considering whether the inductive learning mechanisms that these methods employ have
consequences not simply for natural language engineering, but also for our understanding of the cognitive basis of human
language acquisition and processing. Most machine learning has used supervised learning techniques. These have limited
implications for theories of human language learning, given that they require annotation of the training data with the
structures and rules that are to be learned. However, recently there has been an increasing amount of promising research on
unsupervised machine learning of linguistic knowledge. The results of this research suggest the computational viability of the
view that general cognitive learning and projection mechanisms rather than a richly articulated language faculty may be
sufficient to support language acquistion and intepretation.