Hi Vineet,

Vineet Kashyap schrieb:
> Hi all
>
> Just wanted to know a few things about applying 
> linguistic inputs to improve a baseline SMT system.
>
> 1. Should we tag both the source and the target language 
>    for training? And when we are tuning/testing should that 
>    data be tagged as well?
>   
The necessity to tag both language corpora depends on what you want to 
do with them. If you translate with factors e.g. 0,1,2 - > 0 , then you 
only need factor 0 for the target site. Normally you want to tag both 
language data with all (or almost all) factors to try out different 
combinations of factors. Have a look at:
http://www.statmt.org/moses/?n=Moses.FactoredTutorial
There are a couple of examples with different factor combinations.

The tuning set for MERT should be of the same format as the training 
set. The test set of the foreign language, too. The target language test 
set depends on what translation you want on the output. If it's just the 
word token, you'll need an untagged version of the test set. Moses 
normally tells you if there is a problem with the numbers of factors 
within a data set.
> 2. Also, can language model be tagged and will it make
>    any improvements.
>
> 3. I have the enhanced Brill's tagger for english are there
>    any other good ones?
>
> Finally, when applying linguistic inputs to both source and 
> target language  should they have the same no. of tags like 
> pos, any other morphological information. 
>
>   
You are free to choose the number of factors. It does not need to be the 
same. The essential question is how you want to combine these factors 
because that's what you need to tell Moses by the 
train-factored-phrase-model.perl parameters.

Best regards,
Christine


> Thanks in advance.
>
> Regards,
>
> Vineet
>
>
>
> _______________________________________________
> 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