Hi,
I will take a look a that. I suppose I did not support or test the type 
of reordering model you use. It should work with msd-bidirectional-fe .
As soon as I have a solution I will let you know.
Best,
Marcin

W dniu 23.08.2012 07:50, Katsuhito Sudoh pisze:
> Hi everyone,
>
> I face an problem in multi-thread decoding with a compact binarized 
> reordering model
> compiled using "processLexicalTableMin", by moses (ver. 2d1cafe).
> (described here: http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc6 )
>
> Decoding process always fails soon after its start without any decoding 
> results.
> This only occurs with multi-thread decoding with the compact reordering model;
> the decoder itself works fine with single-thread decoding
> or with an original binarized reordering model produced by 
> "processLexicalTable".
>
> The reordering model was trained by a standard manner with "train-model.perl",
> with "mslr-bidirectional-fe" reordering option.
>
>
> The following message was printed to stderr when moses failed:
>> Offending hypo = this plate and the other end portions of the 
>> [100000100011100000000]  [total=-43.779] <<-8.000, -9.000, 0.000, -1.104, 
>> 0.000, 0.000, -4.730, 0.000, 0.000, 0.000, 0.000, -29.252, -4.922, -6.625, 
>> -14.099, -13.264, 3.000>>
>
> Further I found a bit strange scores in the output from single-thread run.
>> BEST TRANSLATION: further , the plate supporting member 23 , 
>> ランバーサポートプレート|UNK|UNK|UNK 22 in the longitudinal direction end portion of the 
>> sheet , and the position adjustment of of the present perform , . . . , 
>> member (L) to with a the cooperation 
>> [11111111111111111111111111111111111111]  
>> [total=46627944870416590547798387064832.000] <<-62.000, -41.000, -100.000, 
>> -2.709, -0.981, -5.427, -7.517, -3.445, -3.222, 
>> 155409058250575256304156525199360.000, 17418203209007577149166583808.000, 
>> -214.441, -62.693, -86.992, -62.601, -70.820, 17.998>>
>
> Do anyone also face that problem?
>
> Best,
> Katsuhito
> --
> Katsuhito Sudoh
> NTT Communication Science Laboratories, Japan
> [email protected]
>
>
>
>
>
>
> _______________________________________________
> Moses-support mailing list
> [email protected]
> http://mailman.mit.edu/mailman/listinfo/moses-support
>


-- 
dr inż. Marcin Junczys-Dowmunt
Uniwersytet im. Adama Mickiewicza
Wydział Matematyki i Informatyki
ul. Umultowska 87
61-614 Poznań
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to