Hi Bojar,
I think that for BLEU, the reference of length closest to
the output has to be considered.
Best,
Marcello
Marcello Federico
ITC-irst Trento, Italy
On Dec 15, 2006, at 6:31 AM, bo...@us... wrote:
> Revision: 1071
> http://mosesdecoder.svn.sourceforge.net/mosesdecoder/?
> rev=1071&view=rev
> Author: bojar
> Date: 2006-12-14 21:31:47 -0800 (Thu, 14 Dec 2006)
>
> Log Message:
> -----------
> - Fixed bug with brevity penalty:
> If two reference translations of a sentence were equally
> "close" to the
> hypothesis, the *first* one was taken into account, given the
> order of
> references.
> Now the *shorter* is used, making brevity independent on the
> order of
> references. (Papineni etal are not specific about this, either).
> (Consider the case where the hypothesis is 30 words and there
> are two
> references, one of 28 and one of 32 words.)
> - Fixed usage-behaviour inconsistency:
> usage said that ref.0, ref.1, .. are loaded but it loaded only
> ref.1, ref.2,...
>
> Modified Paths:
> --------------
> trunk/scripts/generic/multi-bleu.perl
>
>
> This was sent by the SourceForge.net collaborative development
> platform, the world's largest Open Source development site.
>
> ----------------------------------------------------------------------
> ---
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to
> share your
> opinions on IT & business topics through brief surveys - and earn cash
> http://www.techsay.com/default.php?
> page=join.php&p=sourceforge&CID=DEVDEV
> _______________________________________________
> Mosesdecoder-commits mailing list
> Mos...@li...
> https://lists.sourceforge.net/lists/listinfo/mosesdecoder-commits
|
