Thanks, Hieu.

After I wrote the question, I realized I should re-phrase but then forgot about it. In the standard processing, train-model.perl trains the translation model and lmplz (or other tool) creates the language model. When adding a neural network language mode, it looks like you're confirming that the NPLM tools train the neural network language model. Hence, my question's reference to train-model.perl was in error.

After separately training the three models: translation, language and neural network language, you edit the moses.ini file to use them together as one SMT model. Once the moses.ini file is configured, you set it as the -config argument for mert-moses.perl. At this point, it looks the existence of the additional model is irrelevant to mert-moses.pl.



On 4/7/2015 1:08 PM, Hieu Hoang wrote:
The NPLM isn't just an implementation of a normal backoff language model like SRILM, IRSTLM, KENLM etc.

It scores strings in a different way and requires it's own training. It's a language model in the sense that it gives a score to a sentence in that language

On 24/03/2015 04:17, Tom Hoar wrote:

I haven't done it, but once you've updated the moses.ini file with the NPLM FF, shouldn't the existing train-model.perl & mert-moses.pl <http://mert-moses.pl> scripts run unchanged?



---- liling tan wrote ----

Dear Moses devs/users,

Is there any working script for training a moses model with NPLM?

Is there any working train-model script for training a MT model with bilingual LM? http://www.statmt.org/moses/?n=FactoredTraining.BuildingLanguageModel#ntoc35

Is there a tutorial on Vector/Neural MT for moses on like http://www.statmt.org/moses/?n=Moses.SyntaxTutorial

Are there any slides/script/hints on explanation on how the bilingual LM works with the moses decoder?

Regards,
Liling


_______________________________________________
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