Hi Mohadeseh,
have you tried to
apt-get update # update the apt-get repository with the info in the
sources.list file
apt-get upgrade # upgrade the system with the packages available in the
apt repository
apt-get -f install # fix the installation, removing broken
Hi,
it was not the main topic of the paper, but in
Log-linear weight optimisation via Bayesian adaptation in statistical
machine translation. Germán Sanchis-Trilles, Francisco Casacuberta.
CoLing 2010. http://www.aclweb.org/anthology/C/C10/C10-2124.pdf
I published some stability results
Hi Per and the rest,
Back in EAMT 2011 we presented a paper about a pretty aggressive pruning
strategy for the phrase table, independently from the test set and without any
loss in translation quality. It basically boils down to keeping only those
phrases that would be used when translating
Hi all,
personally I have an implementation of Koehn's 2004 ACL paper about
statistical sifgnificance tests for MT evaluation. It implements both
stand-alone confidence intervals (sec.5, bootstrap resampling) and
paired bootstrap resampling, if a baseline is given. Right now, it
computes
Indeed, I fully agree with the point about understanding the limits. In
fact, in some multi-reference corpora I have observed variations of more
than 10 BLEU points when computing inter-reference BLEU scores (i.e., one
reference against the other references). However, this issue is much
There are also some works addressing phrase-table size reduction. With
very limited loss in translation quality (if any), it has been shown that
the phrase-table can be pruned by up to 97%. See for instance:
http://www.mt-archive.info/EAMT-2011-Sanchis-Trilles.pdf
much different 2 runs can be.
You might wanna talk to Christian Buck who I think is trying to bash the search
graph
output into something more useful
http://article.gmane.org/gmane.comp.nlp.moses.user/4374/
On 24/10/2011 23:05, Germán Sanchis Trilles wrote:
Hello list,
I was trying
easy to work with, so if you want to
make the options more consistent then please go ahead.
cheers - Barry
On Wednesday 26 Oct 2011 11:34:32 Germán Sanchis Trilles wrote:
Hi Hieu,
thanks for your answer. I guess that, in case the search is not completely
deterministic, I will be better off
Hello list,
I was trying to output a search graph with both the single model scores
and the transition value. While the single model scores are produced by
the option -osgx, the transition value is only present when using the
option -osgx. Since I am not doing this myself, I thought that the
10, 2011 at 7:24 PM, Germán Sanchis Trilles gsanc...@dsic.upv.es
wrote:
Hi Neda,
I would need a bit more information in order to know with for sure what
the
problem is. However, my guess is that you are using the mert-moses.pl
script in some random directory, but the path
Hi Neda,
I would need a bit more information in order to know with for sure what
the problem is. However, my guess is that you are using the mert-moses.pl
script in some random directory, but the path to the moses scripts is not
an absolute path. This is:
Hi Felipe,
not that I know. However, assuming you want to use moses-parallel in
order to take advantage of a multi-core desktop with Ubuntu, installing a
regular queue management system is not that costly, since Torque is
conveniently packaged into the distro. In fact, I run moses-parallel on
,
Germán Sanchis-Trilles
On Wed, 15 Sep 2010, musa ghurab wrote:
hi
I have problem when converting phrase-table.gz to hard disk binary image. i got
the following error:
[m...@ibb]# gzip -cd work/20100914/model/phrase-table.gz | LC_ALL=C sort |
nlp/moses/misc/processPhraseTable -ttable 0 0
Hi Nati,
I'm afraid that what you are asking is an open research (and currently
quite active) problem, namely, how to build incremental translation
models. As far as I know, there are a couple of papers suggesting
directions of how to confront this problem, although such papers are
more
corp.ta.vcb -C corp.en_corp.ta.snt
-CoocurrenceFile corp.cooc
the snt2cooc.out binary is to be found in the GIZA++ compilation directory.
Cheers,
Germán Sanchis-Trilles
Quoting sreeja B.P sreej...@gmail.com:
sir,
while running the Giza++ by the command
./GIZA++ -S corp.en.vcb -T corp.ta.vcb -C
Hi Chaitu.
Maybe you would like to take a look at the following paper:
F.J. Och, H. Ney
A systematic Comparison of Various Statistical Alignment Models
Computational Linguistics, Volume 29, Number 1, March 2003
http://www.aclweb.org/anthology-new/J/J03/J03-1002.pdf
Cheers!
Germán
Hi Panos.
I have never seen this specific error.
That being said, in Ubuntu /bin/sh is a symlink to /bin/dash, which
does not happen to work well with the moses scripts.
You can do two things which, I guess, will help you out:
1.- edit training/cmert-0.5/enhanced-mert and exchange /bin/sh by
Hi all.
My mind suddenly stumbled across a wild doubt...
Does -dl 0 mean monotone, or does it rather mean distortion-limit =
inf (as it did in Pharaoh)?
In the second case, how would you specify a monotonicity constraint? -dl 1?
Best regards,
Germán Sanchis
Hi.
The nan errors do not necessarily mean that the moses.ini will be
corrupt, i.e. the moses.ini file may seem correct, but the translation
model built may be of poor(er than expected) quality. Usually, nan
problems during training mean that the training procedure did not run
completely
Hello Joao.
Yes, you can manually edit the moses.ini and change the language model file.
You do not need to retrain the factored model, but, as you say, you do
need to rerun the MERT (tuning) procedure.
Good luck in you research,
Germán Sanchis
Quoting João Graça [EMAIL PROTECTED]:
Hello,
Dear experts:
I have been trying to specify the lmodel-file via command line, but
have failed.
I have tried several possibilities, playing with the around the file
name, but none worked. Here is what I am doing (as specified in the
Moses tutorial):
$ moses -f moses.ini -lmodel-file 0 0 5
21 matches
Mail list logo
