Sorry for leaving that out:

The machine is a linux i686-m64 with 2 GB of internal memory.
I am training it on the europarl v3 corpus which I downloaded from
http://www.statmt.org/wmt08/shared-task.html

I have wrapped the instructions from the baseline system
http://www.statmt.org/wmt08/baseline.html in a python script.

The tokenizing, lowercasing and filtering work fine and produce the proper
output files...

However when running SRILM to train the LM:

self.home + 'srilm/bin/' + self.chipset + '/ngram-count -order 5
-interpolate -kndiscount -text ' + self.workingdir + 'lm/europarl.lowercased
-lm ' + self.workingdir + 'lm/europarl.lm'
        
It just runs for 24 hours without error messages... After that, my job is
killed by the university computer without producing the proper output file.

Regards



-----Original Message-----
From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of amittai e
axelrod
Sent: vrijdag 2 mei 2008 0:14
To: Sanne Korzec
Cc: [email protected]
Subject: Re: [Moses-support] Training Language model SRILM europarl v3

On Thu, May 1, 2008 at 2:59 PM, Sanne Korzec <[EMAIL PROTECTED]> wrote:
>  I have been installing giza, moses and SRILM and am running the supplied
>  perl scripts. However when I train the language model it takes more than
24
>  hours to train the language model! Is this normal or is something going
>  wrong?

Hi Sanne--

That seems kind of long for common language model builds.
What corpus are you training your language model on,what command are
you using to run it, and what's general kind of computer are you
running stuff on?

Cheers,
~amittai

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to