It may be because training data has 2 consecutive space characters. The
cleaning script should have gotten rid of that. Did you run the cleaning
script?
   clean-corpus-n.perl
You can delete that line



On 11 May 2013 08:11, Wenliang Chen <[email protected]> wrote:

> Hi, all
>
> I used moses-1.0 and tried to run Factored training.I met an error when
> tuning the parameters.
>
> I used word and pos tags as input, an example:
> ''|'' nearly|rb 50|cd people|nns are|vbp thought|vbn to|to have|vb
> died|vbn on|in wednesday|nnp as|in the|dt bangladesh|nnp rifles|nnp (|-lrb-
> bdr|nnp )|-rrb- staged|vbd an|dt armed|jj mutiny|nn ,|, ''|'' mohammad|nnp
> quamrul|nnp islam|nnp said|vbd to|to be|vb over|in pay|nn and|cc
> conditions|nns .|.
>
> The command I used for training is as follows:
> perl ../mosesdecoder-RELEASE-1.0/scripts/training/train-model.perl
> -root-dir train -corpus $1 -f vn -e en -alignment grow-diag-final-and
> -reordering msd-bidirectional-fe -lm 0:5:$2:8 -external-bin-dir
> ~/SMT/Moses/mosesdecoder-RELEASE-1.0/tools --alignment-factors 0,1-0,1
>  --translation-factors 0-0+1-1  --reordering-factors 0-0+1-1
>  --generation-factors 0-1 --decoding-steps t0,g0,t1,g1
>
> A warning got from training:
> WARNING: Your [mapping-steps] require generation steps up to id 2 but you
> defined generation steps 0..1
>
> Then I ran the tuning program, but got an error:
> ScoreProducer: Generation start: 1 end: 3
> ERROR:/home/chenwl/SMT/Moses/fact.ve/train/model/generation.0-1.gz:1:
> expected 2 feature values, but found 0
>
> I checked the file generation.0-1.gz, and found the first line has some
> problem.
> ========generation.0-1.gz==========
>   1.0000000  1.0000000
> buonnhang nnp 1.0000000  0.0000007
> cattles nns 1.0000000  0.0000021
> measles vbz 0.0533333  0.0000163
>
> I dont know how the first line comes.
>
> Best
>
> Wenliang
>
>
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
>


-- 
Hieu Hoang
Research Associate
University of Edinburgh
http://www.hoang.co.uk/hieu
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to