Hi,

I need some help regarding the following error.
My EMS with following settings, *crashes at build-generation stage*.
*
input-factors = word pos
output-factors = word pos
alignment-factors = "word -> word"
translation-factors = "word -> word,pos -> pos"
#reordering-factors = "word -> word"
generation-factors = "word -> pos"
decoding-steps = "t0,t1,g0"*

The error seen in TRAINING_build-generation.1.STDERR file is as follows

*Using SCRIPTS_ROOTDIR: /tools/mosesdecoder-master/scripts
(8) learn generation model @ Wed May  1 23:51:27 IST 2013
(8) [0-1] generate generation table @ Wed May  1 23:51:27 IST 2013
ERROR: Can't write
/home/eilmt/wrk-dir/wrk-jhu-fact-hiero/model/generation-table.1.0-1 at
/tools/mosesdecoder-master/scripts/training/train-model.perl line 1773.*

Thanking you,

On Mon, Apr 22, 2013 at 4:14 PM, Hieu Hoang <[email protected]> wrote:

> You can give a tagged corpus to the EMS, using the format:
>   *word1|POS1 word2|POS2 word3|POS3
> *
> *I think you have to set the variable
>   factorized-stem = [filePath]
> *
> *instead of
>   raw-stem = [filePath]
> *
> *However, when you give the EMS raw-stem, it will tokenize, escape
> special characters, and clean the corpus, before word alignment. If you
> give the EMS factorized-stem, it will assume that the data is already
> tokenized, escaped and cleaned. You must make sure that is the case.
>
> *
> *Also, you must make sure the input sentences you give to the decoder is
> tokenized and escape using the same method as you gold standard data.
>
>
> *
> *
> *
> *
> *
>  *
> *
>
>
> On 22 April 2013 06:15, jayendra rakesh <[email protected]> wrote:
>
>> Hi,
>>
>> I have a gold POS tagged parallel corpus available for usage, which is
>> the format
>>
>> *word1/POS1 word2/POS2 word3/POS3*
>>
>> Is there a way to use the gold corpus directly (and in what specific
>> format should it be used ) from the EMS config file instead of writing
>> intermediate factor generation scripts.
>>
>> Also is it possible to add morphological analysis as factors alongside to
>> the POS tagged corpus, directly to the corpus ?
>>
>> --
>> - Jayendra Rakesh.
>>    BTech CSD.
>>
>> _______________________________________________
>> 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
>
>



Jayendra Rakesh.
BTech CSD.
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to