Hi again,

my last problems with the compact tables was a configuration problem,
but I think this one isn't. I'm working on stuff that relies on word
alignment during decoding, but for some phrase pairs,
targetPhrase->GetAlignTerm() is missing alignment pairs in compact
models (everything is dandy with in-memory tables).

I traced the problem back to the variable "a" in
PhraseTableCreator::EncodeLine(), which stores the alignment. Somewhere
between its being filled with (lines 736-742) and the call to
EncodeAlignment(), alignments go missing.

Specifically, this happens in EncodeTargetPhrasePREnc (lines 639-645). I
have no idea what the purpose of this bit is, so I won't touch it. But
it would be great to have the full alignment information during
decoding.

I can send you a sample phrase table (4kb) to reproduce the problem.

best,
Rico

_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to