Hi,

just add "--first-step 3 --last-step 9" when calling
train-factored-phrase-model.perl

--
Felipe.

El mié, 01-07-2009 a las 13:52 -0430, maria sol ferrer escribió:
> Hi, thanks for all your answers, I have already looked on the internet
> and tried those solutions but cant seem to fix the problem. Now I ran
> the sample english-french corpus through GIZA++ (not through the moses
> script that calls Giza) and it seems to generate the dictionary and
> everything fine, now I would like to know if there is a way to run
> moses from step 3, (after generating the GIZA++ files) because my
> problem seemed to be on step 2... Thanks again
> 
> 2009/6/25 Barry Haddow <[email protected]>
>         Hi Maria
>         
>         There's some discussion of the GIZA buffer overflow problem on
>         the GIZA
>         website.
>         http://code.google.com/p/giza-pp/updates/list
>         
>         The problems seems to arise with recent versions of g++. If
>         you could
>         recompile with an earlier version (g++ 4.1) then it should be
>         ok,
>         
>         regards
>         Barry
>         
>         
>         On Thursday 25 June 2009 04:13, maria sol ferrer wrote:
>         > Hi, thanks for your reply, but I have tried wit a cutoff of
>         40, 20, 10 even
>         > 5 words per line and I am still getting the same error, I
>         have followed the
>         > training steps exactly as described and I cant seem to find
>         the problem.
>         > Please any help would be really appreciated! Thank you
>         >
>         > 2009/6/23 John Burger <[email protected]>
>         >
>         > > maria sol ferrer wrote:
>         > > > Hi, I am going to be  working on my thesis with moses
>         and I a having
>         > > > some problems getting started.. So far I was able to
>         download and
>         > > > install Moses and all its necessary tools successfully,
>         tried the
>         > > > one sentence example and worked fine.. Now I am trying
>         to do the
>         > > > training that is available on the "Moses installation
>         and training
>         > > > run-through" (http://www.statmt.org/moses_steps.html)
>         but when I run
>         > > > moses, during step 2.1a I was getting this error:
>         > > >
>         > > > *** buffer overflow detected
>         ***: /home/puchi/tools/bin/GIZA++
>         > > > terminated
>         > > >
>         > > > now, a few days later without changing anything it seems
>         to go to
>         > > > step 2.1b but I get:
>         > >
>         > > In my experience, the buffer overflow error is due to not
>         sufficiently
>         > > filtering your corpus - GIZA is unhappy with very long
>         segments.  That
>         > > same page describes the "Filter out long sentences" step,
>         with an
>         > > example invocation like this:
>         > >
>         > >   clean-corpus-n.perl ... 1 40
>         > >
>         > > What value are you using for the max length?  I would
>         decrease this
>         > > until GIZA stops throwing the error.
>         > >
>         > > Also, I suspect it only appears to go on to step 2.1b
>         because some
>         > > partial file is laying around - don't be fooled.
>         > >
>         > > - John D. Burger
>         > >   MITRE
>         > >
>         > > _______________________________________________
>         > > Moses-support mailing list
>         > > [email protected]
>         > > http://mailman.mit.edu/mailman/listinfo/moses-support
>         
>         
>         --
>         The University of Edinburgh is a charitable body, registered
>         in
>         Scotland, with registration number SC005336.
>         
> 
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support

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

Reply via email to