WHO:	Peter Wiemer-Hastings, Ph.D. , School of Computer Science, Telecommunication,and Information Systems, DePaul University
TOPIC:	"From Turing to Tutoring: Latent Semantic Analysis as Cognitive Model and Budget Natural Language Understanding"
ABSTRACT:	This talk will center on Latent Semantic Analysis (LSA), a vector-based technique for representing and comparing texts. Following a brief history of LSA and description of the process by which its representations are built, LSA will be discussed as a model for human language learning and representation. Despite the fact that it ignores syntax altogether, LSA has neared or matched human performance on a variety of tasks. For single-sentence texts, however, LSA does not perform well, presumably due (at least in part) to its ignorance of syntax. Research by Dennis et al, Kanejiya et al, and myself has explored augmenting LSA with various types of structural knowledge. This work parallels psychological studies on the effects of structure on similarity judgments. The talk will conclude with descriptions of applications of LSA as an expectation-based natural language understanding mechanism.
WHEN:	2/18/2004 2:00:00 PM
WHERE:	SUNY Buffalo, 280 Park Hall SUNY Buffalo, 280 Park Hall

Events Homepage