Hi Andrei,
You might be interested by this paper in the Machine Translation journal:
What types of word alignment improve statistical machine translation?
http://www.springerlink.com/content/xlugp06254484627/
It happens to contain an extrinsic evaluation (in BLEU score of a
phrase-based SMT system) of some word alignment tools listed in the link
sent by Jörg as well as both word-based and phrase-based discriminative
word aligners.
The phrase-based discriminative aligner (available at
http://code.google.com/p/bia-aligner/) achieves the best results when
its model weights are tuned (according to BLEU score) on the whole
training corpus. In the current implementation this is feasible in
practice for corpora up to about 4 million words.
Patrik
On 20/04/12 16:53, Joerg Tiedemann wrote:
Maybe you can find something useful here:
http://opus.lingfil.uu.se/letsmt-trac/wiki/WordAligner
Jörg
On Fri, Apr 20, 2012 at 2:01 PM, Andrei Popescu-Belis
<[email protected]> wrote:
Dear all,
Do you have any suggestions for software to perform word alignment?
I am looking for a solution to word align several pairs of sentences (one is
an imperfect translation of the other), possibly using knowledge obtained
from a larger parallel corpus.
Here are some ideas that do not seem entirely satisfactory:
1. Build a translation model with GIZA++ over the data to align. But there
is not enough data. Maybe adding the large corpus data? (Slow.)
2. If Moses is used, it can output word alignment info, but our
translations are not necessarily done with Moses.
3. There were research systems for the shared task at the 2003/2005
Parallel Texts workshops, but is any piece of software available?
Thank you and best regards,
Andrei Popescu-Belis
Idiap Research Institute | tel: (41 27) 721 7729
Centre du Parc, CP 592 | fax: (41 27) 721 7712
CH-1920 Martigny | [email protected]
Switzerland | www.idiap.ch/~apbelis
_______________________________________________
Mt-list mailing list
_______________________________________________
Mt-list mailing list
