AI/CLI Talk - Nov 21
Thomas Bartold
only) and on Tuesday, November 21 (all day). His talk will be at 4 PM
on 11/21 as part of the new speaker series on "Computation, Language,
and Information" and will be also part of the AI Lab seminar. An
abstract of his talk is enclosed.
Dan is an assistant professor at the Department of Computer Science
and the Beckman Institute, University of Illinois, Urbana-Champaign.
Professor Roth holds a PhD in Computer Science from Harvard
University. He is a recipient of a best paper award at IJCAI'99 as
well of an NSF CAREER award. His areas of interest are artificial
intelligence, machine learning, natural language processing, databases
and information systems, and theoretical computing.
Please let me know if you would like to meet with him while he is
here.
Note the room change: the talk will be in 1005 EECS which can hold up
to 60 people.
Drago
========================================================================
The Computation, Language, and Information seminar
and The Artificial Intelligence seminar
jointly invite you to a talk:
Learning in Natural Language: Theory and Algorithmic Approaches
Tuesday, November 21
4:00-5:15
1005 EECS
Dan Roth
Dept. of Computer Science and the Beckman Institute
University of Illinois - Urbana/Champaign
http://L2R.cs.uiuc.edu/~danr
In language understanding related tasks inferences heavily depend on
knowledge about the language and the world. I will present research
on a learning centered approach to performing these
``knowledge-intensive inferences''.
First, a learning theory account for the major statistical approaches
to learning in natural language will be presented; it will explain why
these methods work although, typically, the assumptions they are based
on do not hold in the data. I will then show how this coherent view
of when and why learning works in this context helps developing a
better learning centered approach to natural language inferences.
This will be done in the context of the SNoW learning architecture.
The emphasis is on a learning architecture that tolerates data of high
dimensionality, supports relational knowledge representations, allows
the incorporation of additional knowledge into the process and can be
used as a building block in higher level inferences. The approach
will be exemplified with experimental evidence from several language
understanding related tasks.