at a guess, berkeley was added because the output from the berkeley
aligner doesn't need to by symmeterized like those from giza++.
'1-to-1', or 'monotonic', is a good name for the alignments in the
train-recaser, but i think the name really refers to symmetrization, in
which case 'none' would be appropriate.
On 01/01/2012 19:44, Tom Hoar wrote:
Thanks Hieu. Sorry, your name is at the top of the revision I had. So,
I open the question to anyone who can shed some light. I understand
the GIZA++ alignment types aren't relevant to the recaser. However,
EMS calls these "alignment-symmetrization-method" and adds "berleley"
as a new value "name", like so:
alignment-symmetrization-method = grow-diag-final-and
vs.
alignment-symmetrization-method = berkeley
I want to know if the simple one-for-one alignment method used in
train-recaser.perl has its own name, or if one of the existing names
would apply.
Tom
On Sun, 1 Jan 2012 16:27:48 +0700, Hieu Hoang
<[email protected]> wrote:
hiya tom
i didn't write that code but from what i know, it doesn't matter. The
alignment types are to do with how would create the final alignment
from the GIZA++ forward & reverse alignment. Since the reccaser is
given the final alignment, those aren't relevant
On Sun, Jan 1, 2012 at 10:38 AM, Tom Hoar
<[email protected]
<mailto:[email protected]>> wrote:
Hieu,
The train-model.perl script and the manual lists these 7
alignment types: intersect, union, grow, grow-final, grow-diag,
grow-diag-final, grow-diag-final-and, srctotgt, tgttosrc.
The train-recaser.perl script creates its own alignment file
between the cased and lowercased data it creates, and initiates
train-model.perl at step 4 with this alignment file.
Does the recaser's alignment file fit into one of the seven
types, or is it its own type? If it's its own type, does that
type have a common name?
Thanks,
Tom
_______________________________________________
Moses-support mailing list
[email protected] <mailto:[email protected]>
http://mailman.mit.edu/mailman/listinfo/moses-support
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support
