On Fri, 2007-11-30 at 06:33, David Talbot wrote:
> Hi Brianna, Phil,
> 
> My guess is that one of the "count of counts" used for computing  
> discount factors in modified Kneser-Ney is zero. (Modified KN requires  
> information about the number of n-grams that occurred once, twice and  
> three times and given the size of the corpus I wouldn't be surprised  
> if there were no 5-grams that occurred 3 times.)
> 
> A smaller model should work better (or a larger corpus). Try setting  
> -order 3 (or maybe even 2) in the ngram-count command.

Thanks David, that fixed it.

About the data sets:


Training data:
wmt07/training/europarl-v3.fr-en.fr 
wmt07/training/europarl-v3.fr-en.en
Format: parallel files, one sentence per line, utf8.
(I think I read somewhere they should be at least 1000 sentences)

Tuning data:
wmt07/dev/dev2006.fr
wmt07/dev/dev2006.en
Format: same as above??
Size: ?? (compared to training data)

Test data:
wmt07/devtest/devtest2006.fr
wmt07/devtest/devtest2006.en
Format: same as above??
Size: ?? Could be whatever size you like I suppose?

I think http://www.statmt.org/wmt07/baseline.html these instructions are
for translating French -> English only, is that correct?

With the "Train Data" command:
bin/moses-scripts/scripts-YYYYMMDD-HHMM/training/train-factored-phrase-model.perl
 -scripts-root-dir bin/moses-scripts/scripts-YYYYMMDD-HHMM -root-dir 
working-dir -corpus working-dir/corpus/europarl.lowercased -f fr -e en 
-alignment grow-diag-final-and -reordering msd-bidirectional-fe -lm 
0:5:working-dir/lm/europarl.lm:0

What is the significance of the flags -f and -e ?
Are they related to msd-bidirectional-fe ?
Can I just put tg instead of fr and everything else stays the same? (my
training files are raw.tg and raw.en)
 
Is it possible to only download the en/fr files rather than the whole
Europarl corpus? I only want to use it confirm that I get the expected
results and therefore have everything set up correctly.

Thanks,
Brianna
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to