Hi Benjamin, You're translating into Japanese? Perhaps your Japanese reference file is not tokenized? multi-bleu.perl expects the reference translations to have spaces. hyp_len and ref_len refer to token counts. If there are no spaces, then there would just be one token per line in the reference file and ref_len would equal the line count, which it sounds like it does in your case. If the reference file is indeed tokenized, then maybe it uses a non-standard space character that multi-bleu.perl is not recognizing. Kevin
On Fri, Jul 12, 2013 at 9:54 AM, Benjamin Körner < [email protected]> wrote: > Hi there, > > I'm using moses / multi-bleu in order to evaluate a pre-translation > sentence reordering system. > > Trying to score my english - japanese translation with multi-bleu.perl > gives back: > > Use of uninitialized value in division (/) at script/multi-bleu.perl > line 120, <STDIN> line 1160. > > Use of uninitialized value in division (/) at script/multi-bleu.perl > line 120, <STDIN> line 1160. > > Use of uninitialized value in division (/) at script/multi-bleu.perl > line 120, <STDIN> line 1160. > > BLEU = 0.00, 0.0/0.0/0.0/0.0 (BP=1.000, ratio=21.528, hyp_len=24973, > ref_len=1160) > > I noticed that hyp_len and ref_len are far from similar. ref_len > equals the number of lines in both files! How is hyp_len computed? > > Any ideas on this? > > Best, > > Benjamin > > _______________________________________________ > 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
