Hi,
Before training my own grammar with the Joshua, I'm trying to use the
Joshua example grammar (in the joshua/example subdir) with cdec. Given
the same models, weights and similar decoder configurations should
both cdec and Joshua give the same log score when decoding a given
utterance?
I've tried reducing the configurations to a minimum for both decoders,
however I am unable to get the same score or translation output from
cdec. Is there a cdec (or Joshua) parameter or step I need that will
allow me to get the same output from cdec?. Below are the parameters
I'm using.
Thanks
Paul
cdec config and weights
---------------------------------------------
add_pass_through_rules=true
grammar=
example.hiero.tm
formalism=scfg
weights=weights.tuned
input=in.txt
PhraseModel_0 1.066893
PhraseModel_1 0.752247
PhraseModel_2 0.589793
---------------------------------------------
Joshua config
----------------------------------------------
tm_file=example/
example.hiero.tm
tm_format=hiero
glue_file=grammars/hiero.glue
glue_format=hiero
default_non_terminal=X
goalSymbol=S
phrasemodel pt 0 1.066893
phrasemodel pt 1 0.752247
phrasemodel pt 2 0.589793
--------------------------------------------