Dear Afli, mert-moses-new.pl performs a case-sensitive optimization by default. If you would like to make it case-insensitive you have to set up the parameter "--nocase" In order to have a case-sensitive optimization your models have to be case-sensitive, of course.
There are a couple of software (Recaser and Truecaser) built on top of Moses to manage case of monolingual data. Look at this page for details: http://www.statmt.org/moses/?n=Moses.SupportTools mert-moses-new.pl does not manage punctuation. So you have to prepare data (training dev and test) according to your needs, i.e. with English punctuated and Arabic not-punctuated text. For automatically inserting punctuation there are some papers on this subject. We published Cattoni, Roldano, Nicola Bertoldi and Marcello Federico, "Punctuating Confusion Networks for Speech Translation" at Interspeech-2007. best regards, Nicola Bertoldi On 3/10/10 10:12 AM, "afli haithem" <[email protected]> wrote: Hi; I'm Haithem AFLI , master student who develops a speech Translation system with moses decoder in LIG-GETALP, Grenoble-France. I would like to treat ponctuation and case with mert-moses-new.pl <http://mert-moses-new.pl/><http://mert-moses-new.pl/> in order to be case sensitive, I use the following command : ./mert-moses-new.pl <http://mert-moses-new.pl/><http://mert-moses-new.pl/> --working-dir=./mert-new --mertdir=/home/bougares/trunk/mert ../eval_notuned/dev06.ar <http://dev06.ar/><http://dev06.ar/> ref. --nbest=100 /usr/local/bin/moses.2009 --no-filter-phrase-table ./model/moses.ini --lambdas="d:0.6,0-1.2 lm:0.5,0-1 tm:0.2,0-0.5;0.2,0-0.5;0.2,0-0.5;0.2,0-0.5;0.2,0-0.5; w:-1,-1.5-0.5" My english ref. are ponctuated and my arabic dev06.ar <http://dev06.ar/><http://dev06.ar/> are no ponctuated. If you could suggest me some ideas, it would help me a lot. Thank you. AFLI Haithem --------------------------------------------------- Haithem AFLI Université Stendhal Grenoble 3 GETALP - Laboratoire d'Informatique de Grenoble <http://www.liglab.fr/spip.php?page=mot&id_mot=184><http://www.liglab.fr/spip.php?page=mot&id_mot=184> Bâtiment IMAG B 210- 385 avenue de la Bibliothèque 38400 Saint Martin d'Hères - <http://www.liglab.fr/spip.php?page=mot&id_mot=184><http://www.liglab.fr/spip.php?page=mot&id_mot=184> Grenoble - France Tél: + 33 (0) 4 76 63 56 82 Mobile: + 33 (0) 6 31 37 28 93 Mail : [email protected] [email protected] [email protected] ________________________________ Hotmail : un service de messagerie gratuit, fiable et complet Profitez-en <https://signup.live.com/signup.aspx?id=60969> _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
