Barry,
Thanks so much. That did just the trick.
--
Taylor Rose
Machine Translation Intern
Language Intelligence
IRC: Handle: trose
Server: freenode
On Mon, 2012-04-02 at 20:55 +0100, Barry Haddow wrote:
> > Do you think there is a problem with it using giza++ when it expects
> > mgiza?
>
> Hi Taylor
>
> Yes, you need to pass the argument -mgiza to train-model.perl so that it
> knows
> to combine the *.part* files at the end. It will need the merge_alignment.py
> script to do this, which should be in the same directory as GIZA++
>
> cheers - Barry
>
>
> >
> > > Hi Taylor
> > >
> > > Are you running GIZA++ or mgiza? You seem to be launching the GIZA
> > > ++binary but the debug reports:
> > >
> > > Starting MGIZA
> > >
> > > What files does giza actually produce (in ./giza.es-en) ?
> > >
> > > cheers - Barry
> > >
> > > On Monday 02 Apr 2012 19:38:30 Taylor Rose wrote:
> > > > I'm running the following command:
> > > >
> > > >
> > > >
> > > > train-model.perl --root-dir . --e en --f es --corpus corpus/data
> > >
> > > --lm
> > >
> > > > 3:0:/home/trose/server/sandbox/test.BloomMap >& LOG
> > > >
> > > >
> > > >
> > > > and I'm getting the following output:
> > >
> > > https://gist.github.com/2286086
> > >
> > > > The final error I get is "ERROR: Giza did not produce the output
> > > >
> > > > file ./giza.es-en/es-en.A3.final. Is your corpus clean
> > >
> > > (reasonably-sized
> > >
> > > > sentences)? at /usr/local/bin/train-model.perl line 1020."
> > > >
> > > >
> > > >
> > > > I'm thinking that something else is failing and causing Giza to not
> > > >
> > > > finish up but I'm not sure where the problem is. My corpus is
> > >
> > > cleaned
> > >
> > > > with clean-corpus-n.perl and limited to 50 tokens. Anyone have an
> > >
> > > idea?
> >
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support
