y for SMT Reordering Tasks. Sara Stymne,
Jörg Tiedemann, Joakim Nivre. WMT 2014.
Similar experiments for French-English gives the best Bleu scores for
GIZA model 4 and fast_align with grow-diag and grow-diag-final-and.
Some other papers that discusses different alignments, and to some
e
Hi,
You might want to have a look at mgiza, and this HOWTO on how to do
force alignment: http://www.kyloo.net/software/doku.php/mgiza:forcealignment
Best,
Sara
On 03/16/2014 05:49 PM, Nima Pourdamghani wrote:
I need to load my probability tables into Giza++. Currently I am
working with HMM
HI,
Does anyone know of any published results which invesitage the effect of
the size of the tuning data set. I'm primarily interested in relation to
Mert, but other optimization methods would also be interesting,
Best,
Sara
___
Moses-support mailing
Hi,
During training the mkcls word classes for each word are stored in
path-to-working-dir/corpus/en.vcb.classes
path-to-working-dir/corpus/en.vcb.classes.cats
Or change en to the language code you're working on. Based on these
files you can recover the class for each word.
Best,
Sara
On 01
Hi!
I have a modified mert-implementation that takes argument for performing
post-processing of some kind.
It takes two arguments:
--postprocess
turns the processing on
--postprocess_script=STRING
the name of the script that does the postprocessing. it has
to read and wr
Hi Chris,
On 01/27/2012 08:47 PM, Chris Tanner wrote:
Hello everyone,
I'm having trouble understanding the -distortion-file parameters that
are passed into Moses; it's reported that Moses accepts:
> -distortion-file: source factors (0 if table independent of
source), target factors,
Hi!
I got alerted to a bug in the lexical reordering code (thanks
Christoffer Tanner!).
In reordering_classes.cpp (in moses/scripts/training/lexical-reordering):
void Model::createSmoothing(double w) {
scorer->createSmoothing(modelscore->get_scores_fe_prev(), w,
smoothing_prev);
scorer->crea
Hi!
I couldn't get multi-threaded Moses to work with lattice decoding. It
didn't crash, but was extremely slow (several hours per sentence). When
running single-threaded it is much faster (seconds-minutes).
I noticed that it says on the Moses webpage that multi-threaded Moses
for lattices is not
It is also the case that using the
more advanced models does not always improve the results over the
default model.
/Sara
>
> On Thu, Sep 1, 2011 at 4:22 PM, Sara Stymne <mailto:sara.sty...@liu.se>> wrote:
>
> Hi,
>
> I have the same experience, that the lexical
Hi,
I have the same experience, that the lexical reordering model does not
always improve over only using the distance based model. I really think
it depends to a large extent on the language pair and the corpus. Of
course there is also the uncertainty of Mert, so it might be worthwhile
to run sev
Hi,
Did you use the flag:
-decoder-flags "-use-alignment-info -print-alignment-info-in-n-best
-alignment-output-file - -include-alignment-in-n-best"
(or whichever flags you want the decoder to use)
as input to the tuning script? That should hopefully help if you didn't.
/Sara
2011-03-15 12:1
Hi!
I am not aware of a tool, and besides, exactly how to merge is really
dependent on how you split compounds, if you mark them, if you normalize
compound parts, if you use factored translation, et. cetera. For an
overview of merging methods, see my EACL-paper:
http://aclweb.org/anthology-new
Hi!
A common problem with SRILM is that it installs as 32 bits as default
even when compiling on a 64 bit machine. If that's the problem, an easy
way to fix that is to change the machine-type script in srilm/sbin, so
the appropriate MACHINE_TYPE is i686-m64, not i686, and recompile srilm.
The
Hi!
Those error messages are related to lexicalized reordering models, and should
not be triggered in your case, since you don't use such models. Are you running
on old version of Moses? I believe we have fixed a bug a while ago, that led to
problems like yours. If you are, try to get the lates
Hi Maria!
The model msd-bidirectional-e will not work, since you can only
condition lexical reordering models on either the foreign phrase (f) or
both the foreign and the target (fe). Only on the target (e) is not
supported.
There is actually an error on the Moses webpage
(http://www.statmt.o
Hi!
When we installed SRILM and Moses on our 64-bit Ubuntu machine we had
some troubles with getting the machine type right. What solved it in the
end was to hack the machine-type script (found in srilm/sbin), so that
it gave the correct machine type, from i686 to i686-m64:
else if (`uname -m
Hi!
I also tried to do this, and it seems that the output-factors flag does
not work.
However, I found another option, the flag -report-all-factors. If it is
used, all possible output factors are given.
It does not work for n-best lists, however. I looked around in the code,
and in the function
or having two processors?
Do you have other recommendations?
/Sara Stymne
___
Moses-support mailing list
Moses-support@mit.edu
http://mailman.mit.edu/mailman/listinfo/moses-support
18 matches
Mail list logo