There is a sign-up schedule for this event that can be found
at
http://apps.cs.utexas.edu/talkschedules/cgi/list_events.cgi

Type of Talk: UTCS Colloquia/AI

Speaker/Affiliation: Percy Lian

g/Stanford University

Talk Audience: UTCS Faculty, Graduate and
Undergraduate Students, and Outside Interested Parties

Date/Ti

me: Friday, October 21, 2011, 11:00 a.m.

Location: ACES 2.302

Host: Ray Mooney

Talk Title: Learning Compositional

Semantics from Weak Supervision

Talk Abstract:
What is the

total population of the ten largest capitals in the US? Building a system t

o answer free-form questions such as this requires modeling the deep semant

ics of language. But to develop practical, scalable systems, we want to a

void the costly manual annotation of these deep semantic structures and ins

tead learn from just surface-level supervision, e.g., question/answer pai

rs. To this end, we develop a new tree-based semantic representation which
has favorable linguistic and computational properties, along with an algo

rithm that induces this hidden representation. Using our approach, we obta

in significantly higher accuracy on the task of question answering compared
to existing state-of-the-art methods, despite using less supervision.

Speaker Bio:
Percy Liang is currently a post-doc at Google and
will be starting as an assistant professor at Stanford next fall. He obtai

ned his B.S./M.S. from MIT and Ph.D. from UC Berkeley. The general theme of
his research, which spans machine learning and natural language processin

g, is learning richly-structured statistical models from limited supervisi

on, most recently in the context of program induction and natural language
semantics. He has won a best student paper at the International Conference
on Machine Learning in 2008, received the NSF, GAANN, and NDSEG fellows

hips, and is also a 2010 Siebel Scholar.