Hi Irene

You can't use bleu on a single sentence. Since it's normally a product of 1,2, 3 and 4-gram precisions, it will be 0 if there's no 4-gram matches. Try evaluating on a larger set, say several 100 or 1000 sentences,

cheers - Barry

On 24/09/12 16:12, Irene Huang wrote:
Hi, Philipp

Thanks for the "compliments".

I just created two dummy files contains exactly the same English sentence. I just randomly copied some of the data from internet to the file.'

The command I use is

~/mosesdecoder/scripts/generic/multi-bleu.perl -lc true file < translated file

Thank you


On Mon, Sep 24, 2012 at 4:35 AM, Philipp Koehn <[email protected] <mailto:[email protected]>> wrote:

    Hi,

    BLEU scores of 0 are extremely hard to get, so you should be proud.

    More seriously: can you send the files that you tested on and how
    you ran the command?

    -phi

    On Sat, Sep 22, 2012 at 4:29 AM, Irene Huang <[email protected]
    <mailto:[email protected]>> wrote:
    > Thank you for the reply.
    >
    > I have another question.
    >
    > I have trained and tuned on the data but when I tested the data and
    > calculated the bleu store, it gave me 0.
    >
    > I just made two dummy files which contain the exact same
    sentences and run
    > the multi-bleu.perl, it still gave me bleu=0
    >
    > Any idea what happened?
    >
    > Thanks
    >
    >
    > On Thu, Sep 20, 2012 at 6:11 PM, Tom Hoar
    > <[email protected]
    <mailto:[email protected]>> wrote:
    >>
    >> You can also provide a comma-separated list of steps and step
    ranges with
    >> the --do-steps argument (exclusive of the --start-step &
    --last-step
    >> argument), such as:
    >>
    >> --do-steps 1
    >> --do-steps 4-6
    >> --do-steps 1,3-4,7,9
    >> --do-steps 1-9 (default)
    >>
    >>
    >>
    >> On 2012-09-21 01:23, Philipp Koehn wrote:
    >>
    >> Hi,
    >>
    >> yes, you can run selected steps when using the --first-step and
    >> --last-step
    >> switches of train-model.perl.
    >>
    >> The step numbers or given here:
    >> http://www.statmt.org/moses/?n=FactoredTraining.HomePage
    >>
    >> So, in your case, you want to continue with --start-step 4
    >>
    >> -phi
    >>
    >>
    >> -phi
    >>
    >> On Thu, Sep 20, 2012 at 6:46 PM, Irene Huang
    <[email protected] <mailto:[email protected]>> wrote:
    >>
    >> Hi, All I just followed the tutorial and finished training a
    baseline. But
    >> I call the training procedure once and it will finish all the
    components
    >> training for me. So I am wondering can these training steps be
    separated?
    >> Like if I have my own alignment model and I give the input as
    moses needed,
    >> like the format in " aligned.grow-diag-final-and", am i able to
    call moses
    >> to continue the phrase extract process and reordering process?
    Thank you
    >> _______________________________________________ Moses-support
    mailing list
    >> [email protected] <mailto:[email protected]>
    http://mailman.mit.edu/mailman/listinfo/moses-support
    >>
    >> _______________________________________________
    >> Moses-support mailing list
    >> [email protected] <mailto:[email protected]>
    >> http://mailman.mit.edu/mailman/listinfo/moses-support
    >>
    >>
    >> _______________________________________________
    >> Moses-support mailing list
    >> [email protected] <mailto:[email protected]>
    >> http://mailman.mit.edu/mailman/listinfo/moses-support
    >>
    >




_______________________________________________
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

Reply via email to