Hello,

I have some problems with running mert when the SL input is a lattice.

1) Weight-i
The transition probabilities in my lattices should not have an impact on
the probability of a generated hypothesis. How can I tell mert that
"weight-i" should not be tuned?

2) Mert aborts with an error message:
I ran mert with a SL input in a lattice format using the following command:

/scripts-20110203-1116/training/mert-moses.pl --no-filter-phrase-table
--decoder-flags '-inputtype 2 -weight-i 0' en_lattice dev_de
moses-cmd/src/moses model/moses_lattice.ini --working-dir tuning_lattice
--rootdir scripts-20110203-1116/

Running this command leads to the following error:

/Can't use an undefined value as an ARRAY reference at
scripts-20110203-1116/training/mert-moses.pl line 688./

Mert seems not to find "I" (weight-i?) although I mention it explicitly
when executing mert. Does my command contain an error?

I also specify "--no-filter-phrase-table" because the filtering seems
not to be successful. I noticed that filtering a binary phrase table
does not produce filtered phrase table (no corresponding file in the
"filtered" subdirectory). Is this intended when having a binary phrase
table? Furthermore, filtering the reordering table does not find any
appropriate entries which is very strange.

/Executing: scripts-20110203-1116/training/filter-model-given-input.pl
./filtered en_lattice
filtering the phrase tables... Mo 21. Mär 12:57:32 CET 2011
Executing: mkdir -p tuning_lattice/filtered
Considering factor 0
filtering model/reordering-table.wbe-msd-bidirectional-fe.gz ->
tuning_lattice/filtered/reordering-table.wbe-msd-bidirectional-fe...
0 of 604634 phrases pairs used (0.00%) - note: max length 10
/

Thanks in advance for your help!


Best,

Anita

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

Reply via email to