I think m_ngram->Order() for some reason returns the n-gram order as a char, 
which is interpreted as a code point rather than a number by the output stream. 
I suppose the solution is to use static_cast<int>(m_ngram->Order()) instead.

/Christian

On Jan 13, 2012, at 12:14 PM, Barry Haddow wrote:

> Hi Joerg
> 
> I looked at my features.list and I see a ctrl-E after 'LM_'. This doesn't 
> split the lines, and so it works.
> 
> Are you using KenLM?
> 
> I tracked the construction of the description down to these lines in 
> LM/Ken.cpp
> 
> std::string GetScoreProducerDescription(unsigned) const {
>     std::ostringstream oss;
>     oss << "LM_" << m_ngram->Order() << "gram";
>     return oss.str();
>   }
> 
> m_ngram->Order() is in kenlm, I think. Kenneth?
> 
> cheers - Barry
> 
> On Friday 13 Jan 2012 10:44:19 Joerg Tiedemann wrote:
>> I get a strange line break (after LM_) with 'moses -show-weights'
>> 
>> which causes my mert process to crash:
>>> moses ....  -show-weights
>> 
>> ...
>> 
>> Distortion d 0.153558
>> LM_
>> gram lm 0.208468
>> WordPenalty w 0.023863
>> InputScore I 0.000000
>> PhraseModel tm 0.146046
>> PhraseModel tm 0.067487
>> PhraseModel tm 0.198544
>> PhraseModel tm 0.004969
>> PhraseModel tm 0.133249
>> PhraseModel tm -0.018513
>> PhraseModel tm 0.034590
>> 
>> 
>> I had to fix the features.list by hand and after that it seems to work.
>> Does anyone know this problem? (and what to do about it?)
>> 
>> Jörg
>> 
> 
> _______________________________________________
> 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