Hi,

I used a former version of Moses for training factored models without
problems (with the script train-factored-phrase-model.
perl), but now I have some problems with the new version (script
train-model.perl) when I use all the factors available. For instance, I have
a source language with 4 factors (extension .cat) and a target language with
2 factors (extension .lsc). When I set the parameter "-translation-factors
0,1,2,3-0", the error is:

(4) generate lexical translation table 0,1,2,3-0 @ Tue Aug 24 09:41:59 CEST
2010
(4) [0,1,2,3-0] generate lexical translation table @ Tue Aug 24 09:41:59
CEST 2010
(1.0.5) reducing factors to produce proves/meteo12/model/aligned.0,1,2,3.cat
@ Tue Aug 24 09:41:59 CEST 2010
Executing: ln -s
proves/corpus/meteo2_factor.catproves/meteo12/model/aligned.0,1,2,
3.cat
(1.0.5) reducing factors to produce proves/meteo12/model/aligned.0.lsc  @
Tue Aug 24 09:41:59 CEST 2010

(proves/meteo12/model/aligned.0,1,2,3.cat
,proves/meteo12/model/aligned.0.lsc,proves/meteo12/model/lex.0,1,2,3-0)
FILE: proves/meteo12/model/aligned.0.lsc
FILE: proves/meteo12/model/aligned.0,1,2,3.cat
ERROR: Can't read proves/meteo12/model/aligned.0,1,2,3.cat at
tools/moses-scripts/scripts-20100823-1022/training/train-model.perl line
1069.


When the parameter is "-translation-factors 0,1,2-0,1", the error is:

(4) generate lexical translation table 0,1,2-0,1 @ Tue Aug 24 09:43:11 CEST
2010
(4) [0,1,2-0,1] generate lexical translation table @ Tue Aug 24 09:43:11
CEST 2010
(1.0.5) reducing factors to produce proves/meteo12/model/aligned.0,1,2.cat
@ Tue Aug 24 09:43:11 CEST 2010

(1.0.5) reducing factors to produce proves/meteo12/model/aligned.0,1.lsc  @
Tue Aug 24 09:43:11 CEST 2010
Executing: ln -s proves/corpus/meteo2_factor.lsc
proves/meteo12/model/aligned.0,1.lsc
(proves/meteo12/model/aligned.0,1,2.cat
,proves/meteo12/model/aligned.0,1.lsc,proves/meteo12/model/lex.0,1,2-0,1)
FILE: proves/meteo12/model/aligned.0,1.lsc
ERROR: Can't read proves/meteo12/model/aligned.0,1.lsc at
tools/moses-scripts/scripts-20100823-1022/training/train-model.perl line
1068.


When I don't use all the factors in any language it works properly. And
there is the same problem with the parameter "-alignment-factors".

I think there is a bug in train-model.perl, subroutine reduce_factors, but I
don't know how to solve it. Could somebody help me?

Thanks in advance,
Guillem
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to