Type of Talk: UTCS Colloquium/AI

Speaker/Affiliation: Ted

Pedersen/University of Minnesota, Duluth

Date/Time: Friday, October

8, 2010, 11:00 a.m.

Location: ACES 2.402

Host: Raymond Mooney

n
Talk Title: The Effect of Different Context Representations on Word Sen

se Discrimination in Biomedical Texts

Talk Abstract: Unsupervised word
sense discrimination relies on the idea that words that occur in similar c

ontexts will have similar meanings. These techniques cluster multiple conte

xts in which an ambiguous word occurs, and the number of clusters discover

ed indicates the number of senses in which the ambiguous word is used. One

important distinction
among these methods is the underlying means of repr

esenting the contexts to be clustered. In this talk I will compare the effi

cacy of first--order methods that directly represent the features that occu

r in a context with several second--order methods that use a more indirect

representation. I will show that second order methods that use word by word
co--occurrence matrices result in the highest accuracy and most robust wor

d sense discrimination. These experiments were conducted with the freely av

ailable open--source software package SenseClusters, using experimental da