Time: 4pm, November 21st
Place: Room 110, South Hall

     Modeling human verbal meaning with Singular Value Decomposition 
                      (SVD) of text corpora.
                        Thomas K. Landauer,
                 University of Colorado at Boulder

Latent Semantic Analysis (LSA) is  computational model of human acquisition
and representation of the meaning of words and passages. It assumes that the
meaning of a passage can be approximated as the sum of the meaning of its
words. Thus, a body of text can be cast as a large system of
(ill-conditioned) simultaneous linear equations, which can be solved by SVD.
The surprising result is that a wide variety of linguistic and
psycholinguistic phenomena are closely modeled in this way. Examples include
simulation of the astounding 10/day rate of vocabulary learning by
school-children, and  passing final exams multiple choice tests.  Moreover,
model simulations have  been used successfully to replace humans in tasks
requiring understanding of text. Perhaps the most dramatic example of the
latter is scoring essay exams as reliably as expert human raters.

Thomas Landauer Bio.

President, Knowledge Analysis Technologies and Professor, University of Colorado 
at Boulder. Worked on applications of machine understanding of ordinary text for
educational and personnel applications based on basic scientific research in
psychology and cognitive science at Bellcore and the University of Colorado. Prof.
Landauer is the author of "The Trouble with Computers: Usefulness, Usability and
Productivity", MIT Press, winner of the 1995 Association of American Publishers 
Best Book in Computer Science Award. He is a Fellow of American Association for 
the Advancement of Science, American Psychological Association, American
Psychological Society, and the Society of Experimental Psychologists.