Moses-support Digest, Vol 92, Issue 44

Send Moses-support mailing list submissions to
moses-support@mit.edu

To subscribe or unsubscribe via the World Wide Web, visit
http://mailman.mit.edu/mailman/listinfo/moses-support
or, via email, send a message with subject or body 'help' to
moses-support-request@mit.edu

You can reach the person managing the list at
moses-support-owner@mit.edu

When replying, please edit your Subject line so it is more specific
than "Re: Contents of Moses-support digest..."


Today's Topics:

1. Re: Recasing is very slow compared to the actual translation
(Stanislav Ku??k)


----------------------------------------------------------------------

Message: 1
Date: Wed, 25 Jun 2014 15:12:19 +0000 (UTC)
From: Stanislav Ku??k <standa.kurik@gmail.com>
Subject: Re: [Moses-support] Recasing is very slow compared to the
actual translation
To: moses-support@mit.edu
Message-ID: <loom.20140625T170622-161@post.gmane.org>
Content-Type: text/plain; charset=us-ascii

>
> the recaser should be very fast. You should set the distortion limit to 0.
>
> you should binarize your phrase table and language model
>

Thanks,

binarizing both phrase table and the language model for the recaser cut the
time to about a half.

I also realized that unlike the actual translation, the recasing phase
probably does not like a high parallelization as using '-th 24' yields much
worse results than '-th 1'

As for the distortion limit, I don't think I can train the recaser to set it
to 0 since I found the following in train-model.perl:

if ($_HIERARCHICAL) {
...
}
else {
print INI "[distortion-limit]\n6\n";
}

So I can rewrite the value manually after the recaser is trained, but
shouldn't it be possible to be able to specify this *before* the training?



------------------------------

_______________________________________________
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support


End of Moses-support Digest, Vol 92, Issue 44
*********************************************

Tweet