Natural language research announcements

[Margaret Fleck <mfleck AT cs.uiuc.edu>]

-------- Original Message --------
Subject: FW: COMPUTER SCIENCE COLLOQ, Michel Galley, May 8 at 10:00 am
Date: Wed, 2 May 2007 14:13:31 -0500
From: Erna A Amerman 
<erna AT cs.uiuc.edu>
To: 
<ifaculty AT cs.uiuc.edu>
CC: 
<clerical AT cs.uiuc.edu>,
 "Vicky L Gress" 
<gress AT uiuc.edu>

Michel Galley is a faculty candidate.
David Forsyth is the technical host assisted by Ronda Pellegrini at
rpellegr AT cs.uiuc.edu

Erna


-----Original Message-----
From: 
announce-bounces AT cs.uiuc.edu
 
[mailto:announce-bounces AT cs.uiuc.edu]
On Behalf Of Erna Amerman
Sent: Wednesday, May 02, 2007 2:07 PM
To: 
announce AT cs.uiuc.edu;
 
cs-grads AT cs.uiuc.edu
Subject: COMPUTER SCIENCE COLLOQ, Michel Galley, May 8 at 10:00 am

University of Illinois at Urbana-Champaign
Department of Computer Science
The Thomas M. Siebel Center for Computer Science
201 North Goodwin Avenue
Urbana, Illinois 61801-2302  USA


COMPUTER SCIENCE COLLOQUIUM


   Statistical Syntax-Directed Text-to-Text Generation


     Michel Galley
     Computer Science Department
     Columbia University
     New York, New York
                                
     May 8 (Tuesday), 2007 at 10:00 a.m.
     2405 Siebel Center for Computer Science
                                

The need for structural and syntactic mappings between languages is
widely recognized in the field of Natural Language Processing (NLP), and

formalisms such as synchronous and transduction grammars have become
increasingly popular. In this talk, we introduce the learning of
weighted synchronous grammars from text corpora of source and target
languages and consider two real-world applications: statistical machine
translation (SMT) and document summarization.

While much work on synchronous grammar induction has been limited to
isomorphic tree mappings, we consider more expressive alternatives that
can account for free transformations often seen in naturally occurring
data. In the case of SMT, we overcome syntactic divergences between
source and target trees by learning non-isomorphic tree mappings using a

linear-time grammar learning algorithm that scales to hundreds of
millions of words. In the case of document summarization, we present how

to learn syntactic compression grammars from unrestricted document and
summary pairs. We show that the use of lexical dependencies and
tree-internal annotation is critical in the production of good sentence
compressions, in particular in determining whether each syntactic phrase

is grammatically optional or mandatory. We present state-of-the-art
performance on both machine translation and summarization.


Bio:
Michel Galley is a Ph.D. candidate in Computer Science at Columbia
University, and a frequent visitor at the University of Southern
California's Information Sciences Institute (ISI). He received his
diploma from the Swiss Federal Institute of Technology (EPFL), and then
worked for one year on statistical natural language generation for
dialog systems at Bell Laboratories. His research is focused on
empirical approaches to natural language processing, with an emphasis on

summarization and machine translation.


If you would like to subscribe to announcement:
http://mail.cs.uiuc.edu/mailman/listinfo/announce

Attachment: POSTER May 8, Michel Galley (Forsyth-RP).doc
Description: MS-Word document