pretty hard
> to beat the phrase-based BLEU for many language pairs.
> http://www.mail-archive.com/moses-support@mit.edu/msg01995.html
> here's Edinburgh's attempt from this years WMT10:
> http://aclweb.org/anthology-new/W/W10/W10-1715.pdf
>
> The straightforward way o
this sounds like an excellent idea. i played around with it a few
years back, but didn't really get around to doing anything much with
it.
you will need to modify experiment.perl. it should be fairly easy to do
Miles
On 31 August 2010 21:36, John Morgan wrote:
> --
> Regards,
> John J Morgan
--
Regards,
John J Morgan
Hello,
I have some questions about using a different word alignment program
in the moses ems.
Has anyone incorporated mttk into the ems?
Would it make sense to do this?
I have the mttk tutorial running on a cluster using qsub so I was
thinking it might be easy to inco
Hi,
sorry - I misunderstood the error message.
The problem is instead that your version of GIZA++ does not
support cooc files. You need to compile it with
"-DBINARY_SEARCH_FOR_TTABLE"
More info on that is here:
http://fjoch.com/GIZA++.html
-phi
2010/8/31 Sonja PETROVIĆ LUNDBERG :
> But the co
But the cooc file is created, it is there.
Sonja
2010/8/31 Philipp Koehn :
> Hi,
>
> since the cooc file was not created, the problem is very likely
> the path to the snt2cooc binary.
>
> -phi
>
> 2010/8/31 Sonja PETROVIĆ LUNDBERG :
>> It was path to mkcls that caused the problem! Now that's fixe
Hi,
since the cooc file was not created, the problem is very likely
the path to the snt2cooc binary.
-phi
2010/8/31 Sonja PETROVIĆ LUNDBERG :
> It was path to mkcls that caused the problem! Now that's fixed, bur not this:
>
> executing
> /Users/so/work/experiments/test/steps/1/TRAINING_r
Hi Joo-Young,the -T option will give you a partial description of the 1-best translation. It doesn't include source non-terminals, so you can't know exactly which rules were applied, but it does show the source spans that were covered by each rule and full target side information. It's described
It was path to mkcls that caused the problem! Now that's fixed, bur not this:
executing
/Users/so/work/experiments/test/steps/1/TRAINING_run-giza-inverse.1
via sh (1)
executing /Users/so/work/experiments/test/steps/1/TRAINING_run-giza.1
via sh (2)
step TRAINING:run-giza-inverse cr
Hi,
you need to look at steps/1/TRAINING_prepare-data.1.STDERR to see what
crashed the step. I presume it has something to do with the installation
path to mkcls. Otherwise, please also check if the input files were created
correctly.
-phi
2010/8/31 Sonja PETROVIĆ LUNDBERG :
> Thanks!
>
> Next p
Dear Wenlong,
The Moses toolkit is language independent, so there shouldn't be anything
special to do. The one thing to take care of is to tokenize properly the
Chinese training corpus. Moses takes as input sentences where words (tokens)
are space-separated, and usually in Chinese texts, words
Dear Wenlong,
The Moses toolkit is language independent, so there shouldn't be anything
special to do. The one thing to take care of is to tokenize properly the
Chinese training corpus. Moses takes as input sentences where words (tokens)
are space-separated, and usually in Chinese texts, words
Thanks!
Next problem happens in the first step:
step TRAINING:prepare-data crashed
number of steps doable or running: 0
I tried rm steps/1/TRAINING_prepare-data.1* and experiment.perl
-continue 1 -exec, but it crashed again, at the same place.
Sonja
2010/8/31 Philipp Koehn :
> Hi,
>
> ghostvie
Hi all,
I use moses-chart and it works well.
But, I want to know and get the translation rules which are used to
translate a given source sentence in decoding time.
Simply said, I try to find a way to know which translation rules are
selected in each ChartCell of moses-chart.
Is there any metho
Hi,
ghostview seems to sometimes act funny and not display
a just recently written file. In this case, you need to manually
type
% gv steps/0/graph.0.ps
after running experiment.perl
-phi
2010/8/31 Sonja PETROVIĆ LUNDBERG :
> weight.ini was missing in that directory, but I've found it in
> an
hi arda
I think an email by Chris Dyer sums up the issue that it's pretty hard
to beat the phrase-based BLEU for many language pairs.
http://www.mail-archive.com/moses-support@mit.edu/msg01995.html
here's Edinburgh's attempt from this years WMT10:
http://aclweb.org/anthology-new/W/W10/W10-1715
weight.ini was missing in that directory, but I've found it in
another, almost identical, directory (with Moses decoder stuff). I
have no idea why there were two different EMS directories, and why
there was no weight.ini in the first one.
Now I experience another problem with the same command "per
Hi everyone,
I just wanted to ask your general opinion about the different systems on
English-German pair and the tree based models.
I was translating a test set with 4 different systems, namely, phrase based,
hierarchical phrase-based, string to tree and tree to tree.
I was hoping to maybe impro
Hi,
the directory
/Users/so/tools/moses-scripts/scripts-20100806-1525/ems/example/data
should contain:
nc-5k.en
nc-5k.fr
test-ref.en.sgm
test-src.fr.sgm
weight.ini
At least that is what is in the SVN directory moses/scripts/ems/example/data.
-phi
2010/8/31 Sonja PETROVIĆ LUNDBERG :
> Hi!
>
>
Hi,
Is there anybody can share your experience about how to train a Moses engine
from Engish to Simplified Chinese?
I know there are some differences and some the training/tuning scripts are
not supporting this language pair as default.
I intend to train one Moses engine for this langauge pair, c
Hi!
I am trying to learn how to use EMS on my computer, but already in the
testing phase, using config.toy that comes with the installation of
Moses, I experience this problem:
ikso-ho:test so$ perl experiment.perl -config config.toy
STARTING UP AS PROCESS 41208 ON ikso-ho.lan AT Tue Aug 31 11:05
20 matches
Mail list logo
