Ok, then the parameters I sent you should work. In your first mail, you
wrote this:
target pos -> target words (generation table 0-1)
The factors are swapped here (0-1 instead of 1-0), so I just wasn't sure
which direction was correct.
Best,
Ales
On 04/10/2013 11:43 AM, Sriram venkatapathy wrote:
Thanks Ales ! It works. Yes, I think I had misunderstood the mapping
section. I am still not clear about the step index that you mentioned,
shall check it out.
The generation table is indeed from part of speech tags to words. Here
is a snapshot,
conj yet 0.0001612 0.1165049
conj both 0.0029155 0.1998158
conj since 0.0031305 0.2045654
conj that 0.0624614 0.5230648
conj if 0.0111111 1.0000000
conj whilst 0.0002150 1.0000000
conj like 0.0002956 0.0161290
conj or 0.0949617 0.9970377
conj soon 0.0001881 0.0474576
Best Regards,
Sriram
On Fri, Apr 5, 2013 at 11:28 AM, Ales Tamchyna <[email protected]
<mailto:[email protected]>> wrote:
Hello Sriram,
I think you misunderstand the [mapping] section of moses.ini (I
don't know if it's documented anywhere though, apart from the
source code).
The first number specifies the decoding path, the second number is
the step index. Your configuration corresponds to this:
[mapping]
0 T 0
0 T 1
0 G 0
If you pass these parameters to train-model.perl, it should
generate the correct moses.ini:
--translation-factors 0-0+1-1 --generation-factors 1-0
--decoding-steps t0,t1,g0
(Assuming you want the generation table from PoS to words, not the
other way around.)
Best,
Ales
On 04/04/2013 05:05 PM, Sriram venkatapathy wrote:
Hello,
I have trained a factored model that maps source syntax to target
syntax (part of speech tags). So, the mappings that I have are,
source word -> target words (phrase table - factors 0-0)
source pos -> target pos (phrase table - factors 1-1)
target pos -> target words (generation table 0-1)
Anyone has idea of what should be order of decoding steps ? This
seemed the right order
0 T 0
0 G 1
1 T 1
but this doesn't stop processing when I give as input a sentence
tagged with part of speech tags. Any other mapping order leads to
a segmentation fault.
Any suggestions about this ? Thanks.
- Sriram
_______________________________________________
Moses-support mailing list
[email protected] <mailto:[email protected]>
http://mailman.mit.edu/mailman/listinfo/moses-support
_______________________________________________
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
