UTCS Colloquium/AI: Miles Osborne/University of Edinburgh Discriminative Synchronous Transduction for Statistical Machine Translation ACES 2.302 Monday March 31 2008 3:00 p.m.

Contact Name: 
Jenna Whitney
Mar 31, 2008 3:00pm - 4:00pm

There is a sign up schedule for this event:

Type of Talk: UTCS Colloquium/AI

Speaker/Affiliation: Miles

Osborne/University of Edinburgh

Date/Time: Monday March 31 2008

3:00 p.m.

Location: ACES 2.302

Host: Raymond Mooney

Talk Title:
Discriminative Synchronous Transduction for Statistical Mach

ine Translation

Talk Abstract:
Large-scale discriminative machine
translation promises to further
the state-of-the-art but has failed to
deliver convincing gains over
current (mainly) generative systems. We

argue that a principle
reason for this failure is not dealing with mult

iple equivalent ways
of producing the same translation for a given sou

rce sentence. We
present a translation model based upon a synchronous c

ontext free
grammar. By explicitly summing-out all possible derivations
we are
able to improve translation results. Estimation is exact sinc

e we can
employ dynamic programming techniques used in the log-linear <

br>parsing literature when calculating expectations. Decoding however

s only approximate since no tractable algorithms exist for marginal-

ing-out derivations.

Joint work with Phil Blunsom and Trevor Cohn.
Speaker Bio:
Miles Osborne is on the faculty at Informatics Edinb

urgh University.
His main interests are in machine learning applied to

natural language
statistical machine translation and more recently ra

ndomised algorithms.
He has started looking at processing Blogs using H

adoop a handful of
derelict machines and some cardboard boxes.