Yes, giza++ has a support for an optional dictionary file. You can read more
about it in giza++ 's README file.

If you mean an option in train-factored-phrase-model to give a dictionary to
giza++, then it seems a little complicated, because the dictionary file
needs to contain word indexes, which are made in the first phase of the
training. It's probably possible to start the first phase, then build your
dictionary file using the vocabulary index files made, and resume the
training at phase 2, adding "-giza-option=d=<dictfile>". Or to patch
train-factored-phrase-model.perl.

Personally, I found it more simple to just consider dictionary entries as
one-word sentences that I embed in the starting corpus.

There is a difference: in giza++'s first iteration, when there is a
"dictionary match" inside 2 aligned sentences, it won't try to align any of
the 2 matched words with any word other than its match in the other
sentence. I'm not sure of how useful this is.

If you do "properly" integrate a dictionary, I'd be interested to hear how
you've done.

Best regards,

-- 
Raphael Payen

 
From: [email protected] [mailto:[email protected]]
On Behalf Of rashlin ..
Sent: 14 April 2010 15:40
To: [email protected]
Subject: [Moses-support] dictionary help

> hello all....
>
>     Is there any option for adding dictionary in GIZA++???....
>
> thanking you
> rashlin..



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

Reply via email to