[Colloquium] Talk Reminder: Today at 11:00a.m.

[Dawn Ellis]

When:  Friday, April 19th at 11a.m.

Where: TTIC-6045 S. Kenwood Ave.  Room 526

Speaker: Vivek Srikumar (UIUC)

Title: Structured Learning and Prediction for Semantic Analysis of Text

Abstract:

Much of the world's knowledge is textual. The ability to automatically
convert text into a representation that captures its meaning is
crucial to developing intelligent, language-driven applications.  I
will present research on statistical machine learning protocols and
inference algorithms that support wh-analysis of text at the level of
"who does what to whom, when and where".

Analyzing text at this level is a structured prediction task, in that
it requires making multiple inter-related decisions. Applying standard
supervised learning techniques to this and related problems is
hindered by the difficulty of annotating large amounts of text at this
level. I will discuss work that shows how to exploit interactions
among components of the desired output along with linguistic knowledge
in order to train good statistical models in the presence of partial
or even missing annotation. In particular, these techniques allow us
to extend previous work on this wh-analysis for verb predicates (also
called semantic role labeling) to support analysis with respect to
additional predicates without requiring full supervision.  Finally, I
will address the computational aspect of predicting these
predicate-argument representations and show that by taking advantage
of regularities of the output over the lifetime of the predictor, we
can obtain a significant amortized gain in running time without
compromising on accuracy.

Host:  Karen Livescu, klivescu at ttic.edu

-- 
*Dawn Ellis*
Administrative Assistant
773-834-1757
dellis at ttic.edu

TTIC
6045 S. Kenwood Ave.
Chicago, IL. 60637
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20130419/5fef2418/attachment.htm