r strides in recent years. An important enabling
factor has been the ad

option of automatic evaluation
metrics to guide researchers in making i

mprovements
to their systems. Research in automatic evaluation
metr

ics faces two major challenges. One is to achieve
higher agreement with
human judgments when
evaluating MT outputs at the sentence-level. Anot

her
is to minimize the reliance on expensive human-developed
resou

rces such as reference sentences. In this talk I
present a regression-

based approach to metric development.
Our experiments suggest that by c

ombining a wide range
of features the resulting metric has higher corr

elations with
human judgments. Moreover we show that the features do <

br>not have to be extracted from comparisons with human
produced refere

nces. Using weaker indicators of fluency
and adequacy our learned metr

ics rival standard reference-
based metrics in terms of correlations wit

h human judgments
on new test instances.

Speaker Bio:
Rebecca
Hwa is an Assistant Professor in the Department
of Computer Science at
the University of Pittsburgh. Her
research focus is on multilingual pr

ocessing machine
translation and semi-supervised learning methods. Be

fore
joining Pitt she was a postdoctoral research fellow at the
Un

iversity of Maryland. She received her PhD from Harvard
University and