<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
<title></title>
</head>
<body bgcolor="#ffffff" text="#000000">
<p>[apologies for multiple cross-postings]<br>
</p>
<p>Dear NLP-researchers,
<br>
<br>
We're glad to announce the release of the <a
href="http://www.lsi.upc.edu/%7Enlp/IQMT/">IQ<font size="0">MT</font>
Framework for Machine</a><br>
<a href="http://www.lsi.upc.edu/%7Enlp/IQMT/">Translation Evaluation </a><a
href="http://www.lsi.upc.edu/%7Enlp/IQMT/">v 1.0</a>, an open source
software package, released under<br>
the <a href="http://www.fsf.org/licenses/lgpl.html" target="_top">GNU
Lesser</a><a href="http://www.fsf.org/licenses/lgpl.html" target="_top">
General Public License (LGPL)</a> of the<a href="http://www.fsf.org"
target="_top"> Free Software Foundation</a>.<br>
This tool is a joint effort by the <a href="http://nlp.uned.es/">UNED
NLP & IR</a><a href="http://nlp.uned.es/"> Research Group</a> and
the<br>
<a href="http://www.lsi.upc.edu/%7Enlp">TALP Research Center</a><a
href="http://www.lsi.upc.edu/%7Enlp"> NLP group</a>.</p>
<p> The <a href="http://www.lsi.upc.edu/%7Enlp/IQMT/">IQ<font
size="0">MT</font></a> package <a
href="http://www.lsi.upc.edu/%7Ejgimenez/PUBS/IWSLT05-gah.pdf">(Giménez
et al., IWSLT'2005)</a> is based on the QARLA<br>
Framework (<a href="http://acl.ldc.upenn.edu/P/P05/P05-1035.pdf">Amigó
et al.,</a><a href="http://acl.ldc.upenn.edu/P/P05/P05-1035.pdf">
ACL'2005</a>). Rather than defining a new
supermetric<br>
'XXX' our tool follows a 'divide and conquer' strategy.
You can define a set of<br>
metrics and then combine them into a single measure of MT quality, in a
robust<br>
and elegant manner, avoiding scaling problems and
metric weightings.</p>
<p> Using <a href="http://www.lsi.upc.edu/%7Enlp/IQMT/">IQ<font
size="0">MT</font></a> offers a number of advantages over previous MT
evaluation
packages.<br>
First, individual
metrics improve their level of correlation with human judgements<br>
when they are
applied inside QARLA. Second, it permits to avoid the 'metric bias'<br>
problem, by allowing you to tune your system on a combination of
metrics instead<br>
of
on a single metric. Third, it allows you to define a set of subtle
metrics focusing<br>
on partial
aspects of MT quality possibly at different linguistic levels, and then<br>
combine them into a single measure.<br>
</p>
<p> Several well-known and freely available current MT evaluation
tools such as<br>
BLEU, GTM, METEOR, NIST, ROUGE, WER and PER, have been incorporated<br>
so far. We plan to incorporate new metrics working at linguistic levels
other
than<br>
lexical overlap in a near future.<br>
</p>
<p> However, the main feature of the IQ<font size="0">MT</font>
package is that it allows to supply<br>
user-defined MT metrics.
So, do not hesitate to try your own metrics inside<br>
QARLA!<br>
</p>
<p> Please feel free to download IQMT v1.0 at <a
href="http://www.lsi.upc.edu/%7Enlp/IQMT/">http://www.lsi.upc.edu/~nlp/IQMT/</a>
.</p>
<p> Any feedback (comments, suggestions, bug reporting) from the NLP
community<br>
members will be highly appreciated.<br>
</p>
<p><br>
IQMT developing team.<br>
</p>
<p><br>
</p>
</body>
</html>