Massachusetts Institute of Technology
Sign in

"Ideal" Language Learning and the Psychological Resource Problem

10/19/2007 3:50 Wong Auditorium
William Gregory Sakas, Associate Professor, Department of Computer Science, Hunter College
; Ph.D. Programs in Computer Science and Linguistics, The Graduate Center, City University of New York

Description: Some linguists study what can be learned in principle, but William Gregory Sakas asks "the feasibility question -- how efficient learning takes place." This talk focuses on such research, its historical antecedents, and issues that trouble Sakas and his colleagues.

Sakas provides a swift conceptual survey of modeling parameter"setting, from Chomsky, through Yang and Lightfoot. In "the spirit of Pinker," Sakas believes that any computational model of a feasible learner must be compatible with the psychological resources of a human child. So Sakas's lab tries to zero in on "what is needed in the way of psycho"computational resources in a learner to converge on the target grammar on the basis of a limited sample of sentences." The lab has created "a large, artificial but linguistically motivated domain of parameterized languages for evaluating learning models," with more than 1.6 million parsed sentences. Sakas and his colleagues compare the efficiencies of different parameter"setting models, attempt to solve such modeling problems as noise and over"generalization, as well as evaluate richness of the stimulus claims.

Underlying this work _ "the whole parameter setting enterprise" -- says Sakas, is the effort "to limit the resources to what can reasonably be attributed to young children," so as to reduce the complexity of innate knowledge -- generally limit the amount and complexity of input and processing of each sentence. The sticky problem remains of finding a "psycho"computationally palatable way" of modeling a process of fitting grammar to multiple sentences.

"We feel alone in this endeavor," says Sakas. While there's lots of interesting recent work on modeling syntax acquisition, mathematical, and statistical/probabilistic learning, "we haven't been able to take it into our models, because it doesn't seem to be concerned with resource limits." The ideal learner other linguists discuss "is not a learner concerned with resource issues." Sakas asks if "this work is intended to mirror psychological reality." He notes, "The effective richness of the stimulus for a child language learner depends on the child's non"ideal capacity to extract information from heard utterances." The point is to model what a child really does with the linguistic input available to her. Sakas concludes with a question: "Is there anyone out there trying to solve the psychological resource problem?

About the Speaker(s): William Gregory Sakas has been a professor at CUNY/Hunter College since September 2000.

He has published a book chapter with J.D. Fodor in Language Acquisition and Learnabilityand co"authored a textbook, The Core Guide to PPL Programming.

He received his A.B. from Harvard College in Economics and his Ph.D. in Computer Science from the City University of New York.

Host(s): School of Engineering, Laboratory for Information and Decision Systems

Comments (0)

It looks like no one has posted a comment yet. You can be the first!

You need to log in, in order to post comments.

MIT World — special events and lectures

MIT World — special events and lectures

Category: Events | Updated 6 months ago

Created
December 14, 2011 14:29
Category
Tags
License
All Rights Reserved (What is this?)
Additional Files


Viewed
4396 times

More from MIT World — special events and lectures

Introduction and Overview

Introduction and Overview

Added over 4 years ago | 00:39:48 | 2957 views

Jay Gatsby and the Myth of American Origins

Jay Gatsby and the Myth of American...

Added over 4 years ago | 01:25:00 | 4860 views

Software Innovation-Do You Think the Last 20 Years Were Exciting? The Next 20 Years Will Blow Your Mind

Software Innovation-Do You Think th...

Added over 4 years ago | 00:50:12 | 7655 views

Vision: Challenges and Prospects

Vision: Challenges and Prospects

Added over 4 years ago | 01:01:00 | 2539 views

A Beautiful Mind: Genius, Madness, Reawakening

A Beautiful Mind: Genius, Madness, ...

Added over 4 years ago | 01:15:00 | 5998 views

The Future of Science Journalism

The Future of Science Journalism

Added over 4 years ago | 01:42:00 | 6270 views