4 questions

1. Baseline french-to-english system on europarl corpora.
What BLEU scores should whe have with and without tuning stage?

2. tuning stage and MERT algorithm. How long it takes for baseline
system? How many iterations?
What are the best BLEU score on each iteration?

3. we have 7 parameters for distortion model (-d) and 5 parameters for
translation model (-tm) (baseline system). Where this numbers come
from?
   Can I (How can I) reduce the number of this parameters to 1?

4. For a given moses decoder (translation model "f->e", language model
e, distortion model, word penalty)
  and for a given foreign sentence "f", how can I compute (using
mosedecoder) probability (cross-entropy per sentence) for a given
sentence "e", that is translation of sentence "f"?
  For example I have mosedecoder "f->e" and I would like to align
sentences more exactly and remove bad alignments with low probability

Re: mose compile errors with irstlm

Hi Emmanuel,

Looks like the IRSTLM team updated their Moses interface files but  
haven't generated a new official release TGZ file. You should get the  
latest IRSTLM release by SVN checkout instead with the following  
command:

  svn co https://irstlm.svn.sourceforge.net/svnroot/irstlm irstlm

That fixed the problem for me - BUT I had to compile under tcsh on  
linux.

Following from that, I want to report a related problem to the list:

IRSTLM's new ./install script and src/makefile use the $MACHTYPE  
environment variable to create lib/ and bin/ sub-directories. This  
value differs under tcsh ("i386" on my system) and bash ("i686-redhat- 
linux-gnu"). Also, (at least for me) bash fails to pass $MACHTYPE on  
to the call to "make all" within ./install, resulting in slightly  
garbled bin and lib structures if you use bash, and the following  
compilation error:

make[2]: Entering directory `/amd/nfs/har/disk/ptn162/josh/tools- 
latest/moses-test/moses-cmd/src'
g++  -g -O2  -L/group/project/statmt2/josh/tools-latest/irstlm/lib/ 
i686-redhat-linux-gnu -o moses  Main.o mbr.o IOStream.o  
TranslationAnalysis.o -L../../moses/src -lmoses -lirstlm -lz
/usr/bin/ld: cannot find -lirstlm

Short term solution: Check out from svn instead of downloading the  

Re: mose compile errors with irstlm

Hi Emmanuel,

modifications by Josh were checked in. I'm currently
preparing another release including a simpler script
to prepare gigantic LMs. Should be a matter of few
days (still doing regression tests).

Stay tuned, I will make the announcement through
this mailing list.

Greetings,
Marcello

On May 31, 2007, at 11:30 PM, emmanuel adam wrote:

> Hi everybody,
>
> Moses.2004-05-29 doesn't seem to compile with irstlm
> (irstlm-5.00.1.tgz) on my machine (Linux i686). After running make -j
> 4, I get the following errors (compilation with srilm works well):
>
> LanguageModelIRST.cpp: In member function 'virtual bool
> LanguageModelIRST::Load(const std::string&, FactorType, float,
> size_t)':
> LanguageModelIRST.cpp:85: error: invalid use of undefined type  
> 'struct lmmacro'
> LanguageModelIRST.h:35: error: forward declaration of 'struct lmmacro'
> LanguageModelIRST.cpp:105: error: 'class lmtable' has no member  
> named 'getDict'
> LanguageModelIRST.cpp:111: error: 'class lmtable' has no member  

problem installng moses

Re: problem installng moses

Re: problem installng moses

check out josh's email on the same topic
	
http://mailman.mit.edu/pipermail/moses-support/2007-June/000322.html

Hieu Hoang
www.hoang.co.uk/hieu
-----Original Message-----
From: moses-support-bounces@...
[mailto:moses-support-bounces@...] On Behalf Of Hasibul Haque
Sent: 01 June 2007 17:38
To: Moses-support@...
Subject: [Moses-support] problem installng moses

Trying to install moses with irstlm.
getting this errors while running make:
LanguageModelIRST.cpp:29:21: lmmacro.h: No such file or directory
LanguageModelIRST.cpp: In member function `virtual bool
LanguageModelIRST::Load(const std::string&, FactorType, float, size_t)':
LanguageModelIRST.cpp :85: error: invalid use of undefined type `struct
lmmacro'
LanguageModelIRST.h:35: error: forward declaration of `struct lmmacro'
LanguageModelIRST.cpp:105: error: 'class lmtable' has no member named
'getDict' 
LanguageModelIRST.cpp:111: error: 'class lmtable' has no member named
'getDict'
LanguageModelIRST.cpp: In member function `void
LanguageModelIRST::CreateFactors(FactorCollection&)':
LanguageModelIRST.cpp :131: error: 'class lmtable' has no member named
'getDict'
LanguageModelIRST.cpp: In member function `int
LanguageModelIRST::GetLmID(const std::string&) const':
LanguageModelIRST.cpp:169: error: 'class lmtable' has no member named
'getDict' 
make[2]: *** [LanguageModelIRST.o] Error 1
make[2]: Leaving directory `/home/hhaque/mosesdecoder/moses/moses/src'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/hhaque/mosesdecoder/moses' 
make: *** [all] Error 2

baseline, tuning, bleu, parameters, prob

Dear moses experts,

I have some questions, help pls

1. Baseline french-to-english system on europarl corpora.
What BLEU scores should we have with and without tuning stage?
Does tuning stage help us?

2. tuning stage and MERT algorithm. How long it takes for baseline
system? How many iterations?
What are the best BLEU score on each iteration?
I have  0.03, 0.04, 0.11 on the first 3 iterations...

3. we have 7 parameters for distortion model (-d) and 5 parameters for
translation model (-tm) (baseline system). Where this numbers come
from?
  Can I reduce the number of this parameters to 1 (-d) and 1 (-tm) for
a good robustness?

4. For a given moses decoder (translation model "f->e", language model
e, distortion model, word penalty)
 and for a given foreign sentence "f", how can I compute (using
mosedecoder) probability (cross-entropy per sentence) for a given
sentence "e", that is translation of sentence "f"?
  For example I have mosedecoder "f->e" and I would like to align
sentences more exactly and remove bad alignments with low translation
probability.
How can I compute that probability?

Thanx in advance!

http://sz.ru/parser/

Re: baseline, tuning, bleu, parameters, prob

Hi sergey,

Saw you last email. Don't really know the answers, but try my best:

1. you must do parameter tuning, it helps a lot. 
2. mert should take about a dozen iterations. 
3. the 7 distortion paramters - 
		1st - distance based distortion penalty
		next 6 parameters - for lexicalised re-ordering
	u can reduce distortion param to 1 by not doing lexicalised
re-ordering.
   The 5 translation params are detailed here:
	http://www.statmt.org/moses/?n=FactoredTraining.ScorePhrases
   u can reduce this to 1 by only using 1 translation, I suppose the
best to keep would be phi(e|f). After doing the mert, u can reduce it
from 5 to 1 by combining all the params together, since u know have the
optimal weights.

4. not sure what u mean. However, in general, the decoder deals with
scores, or 'feature functions'. U can't convert scores to the
probability. 
U can get the score for any translation, by calling the function
GetTotalScore() in the hypothesis class.

Hieu Hoang
www.hoang.co.uk/hieu

-----Original Message-----
From: moses-support-bounces@...
[mailto:moses-support-bounces@...] On Behalf Of Sergey Protasov
Sent: 01 June 2007 21:00
To: moses-support@...
Subject: [Moses-support] baseline, tuning, bleu, parameters, prob

Dear moses experts,

I have some questions, help pls

1. Baseline french-to-english system on europarl corpora.
What BLEU scores should we have with and without tuning stage? Does
tuning stage help us?

2. tuning stage and MERT algorithm. How long it takes for baseline
system? How many iterations? What are the best BLEU score on each
iteration? I have  0.03, 0.04, 0.11 on the first 3 iterations...

3. we have 7 parameters for distortion model (-d) and 5 parameters for
translation model (-tm) (baseline system). Where this numbers come from?
  Can I reduce the number of this parameters to 1 (-d) and 1 (-tm) for a
good robustness?

4. For a given moses decoder (translation model "f->e", language model
e, distortion model, word penalty)  and for a given foreign sentence
"f", how can I compute (using
mosedecoder) probability (cross-entropy per sentence) for a given
sentence "e", that is translation of sentence "f"?
  For example I have mosedecoder "f->e" and I would like to align
sentences more exactly and remove bad alignments with low translation
probability. How can I compute that probability?

Thanx in advance!

http://sz.ru/parser/ _______________________________________________
Moses-support mailing list
Moses-support@...
http://mailman.mit.edu/mailman/listinfo/moses-support

Re: mose compile errors with irstlm

Hi Josh, hi Marcello,

Thanks a lot for your replies. I got moses running with Josh's workaround.

Greetings, Emmanuel

On 6/1/07, Marcello Federico <federico@...> wrote:
> Hi Emmanuel,
>
> modifications by Josh were checked in. I'm currently
> preparing another release including a simpler script
> to prepare gigantic LMs. Should be a matter of few
> days (still doing regression tests).
>
> Stay tuned, I will make the announcement through
> this mailing list.
>
> Greetings,
> Marcello
>
>
>
>
>
>
> On May 31, 2007, at 11:30 PM, emmanuel adam wrote:
>
> > Hi everybody,
> >
> > Moses.2004-05-29 doesn't seem to compile with irstlm
> > (irstlm-5.00.1.tgz) on my machine (Linux i686). After running make -j
> > 4, I get the following errors (compilation with srilm works well):
> >
> > LanguageModelIRST.cpp: In member function 'virtual bool
> > LanguageModelIRST::Load(const std::string&, FactorType, float,
> > size_t)':
> > LanguageModelIRST.cpp:85: error: invalid use of undefined type
> > 'struct lmmacro'
> > LanguageModelIRST.h:35: error: forward declaration of 'struct lmmacro'
> > LanguageModelIRST.cpp:105: error: 'class lmtable' has no member
> > named 'getDict'
> > LanguageModelIRST.cpp:111: error: 'class lmtable' has no member
> > named 'getDict'
> > LanguageModelIRST.cpp: In member function 'void
> > LanguageModelIRST::CreateFactors(FactorCollection&)':
> > LanguageModelIRST.cpp:131: error: 'class lmtable' has no member
> > named 'getDict'
> > LanguageModelIRST.cpp: In member function 'int
> > LanguageModelIRST::GetLmID(const std::string&) const':
> > LanguageModelIRST.cpp:169: error: 'class lmtable' has no member
> > named 'getDict'
> > make[2]: *** [LanguageModelIRST.o] Erreur 1
> > make[2]: *** Attente des tâches non terminées....
> > make[1]: *** [all-recursive] Erreur 1
> > make: *** [all] Erreur 2
> >
> > Any hint welcome,
> >
> > Emmanuel
> >
> > _______________________________________________
> > Moses-support mailing list
> > Moses-support@...
> > http://mailman.mit.edu/mailman/listinfo/moses-support
>
>
> ------------------
> ITC -> dall'1 marzo 2007 Fondazione Bruno Kessler
> ITC -> since 1 March 2007 Fondazione Bruno Kessler
> ------------------
>

Re: baseline, tuning, bleu, parameters, prob

Thank you, Hieu!

1.
I found that the tuning stage can add only 0.4-3.4 points to the BLEU metric
http://hermes.itc.it/people/bertoldi/JHU-CLSP/met-result.pdf

2. It takes one day for each tuning iteration for baseline system for
my P4-2Gz-2Gb
   So 14-15 days for all tuning stage is very long time.

3. Ok, I will try it

4. For my language pair (english-russian) it is difficult to get
parallel corpora like europarl transcripts.
I have many text (literature,fiction,fairy-tale) that I need to align
and remove all bad pairs.
I need some criteria to sort sentence pairs and remove with low scores.

2007/6/2, Hieu Hoang <h.hoang@...>:
> Hi sergey,
>
> Saw you last email. Don't really know the answers, but try my best:
>
> 1. you must do parameter tuning, it helps a lot.
> 2. mert should take about a dozen iterations.
> 3. the 7 distortion paramters -
>                 1st - distance based distortion penalty
>                 next 6 parameters - for lexicalised re-ordering
>         u can reduce distortion param to 1 by not doing lexicalised
> re-ordering.
>    The 5 translation params are detailed here:
>         http://www.statmt.org/moses/?n=FactoredTraining.ScorePhrases
>    u can reduce this to 1 by only using 1 translation, I suppose the
> best to keep would be phi(e|f). After doing the mert, u can reduce it
> from 5 to 1 by combining all the params together, since u know have the
> optimal weights.
>
> 4. not sure what u mean. However, in general, the decoder deals with
> scores, or 'feature functions'. U can't convert scores to the
> probability.
> U can get the score for any translation, by calling the function
> GetTotalScore() in the hypothesis class.
>
> Hieu Hoang
> www.hoang.co.uk/hieu
>
>
> -----Original Message-----
> From: moses-support-bounces@...
> [mailto:moses-support-bounces@...] On Behalf Of Sergey Protasov
> Sent: 01 June 2007 21:00
> To: moses-support@...
> Subject: [Moses-support] baseline, tuning, bleu, parameters, prob
>
>
> Dear moses experts,
>
> I have some questions, help pls
>
> 1. Baseline french-to-english system on europarl corpora.
> What BLEU scores should we have with and without tuning stage? Does
> tuning stage help us?
>
> 2. tuning stage and MERT algorithm. How long it takes for baseline
> system? How many iterations? What are the best BLEU score on each
> iteration? I have  0.03, 0.04, 0.11 on the first 3 iterations...
>
> 3. we have 7 parameters for distortion model (-d) and 5 parameters for
> translation model (-tm) (baseline system). Where this numbers come from?
>   Can I reduce the number of this parameters to 1 (-d) and 1 (-tm) for a
> good robustness?
>
> 4. For a given moses decoder (translation model "f->e", language model
> e, distortion model, word penalty)  and for a given foreign sentence
> "f", how can I compute (using
> mosedecoder) probability (cross-entropy per sentence) for a given
> sentence "e", that is translation of sentence "f"?
>   For example I have mosedecoder "f->e" and I would like to align
> sentences more exactly and remove bad alignments with low translation
> probability. How can I compute that probability?
>
>
> Thanx in advance!
>
> http://sz.ru/parser/ _______________________________________________
> Moses-support mailing list
> Moses-support@...
> http://mailman.mit.edu/mailman/listinfo/moses-support
>
>