moses-smt/mosesdecoder
1
1
Fork 0
mirror of https://github.com/moses-smt/mosesdecoder.git synced 2025-01-09 04:56:57 +03:00
Code Issues Packages Projects Releases Wiki Activity
1,713 Commits 136 Branches 26 Tags 147 MiB
5b1541b96d
Commit Graph

266 Commits

Author SHA1 Message Date
bojar
5b1541b96d Fixed order of feature scores passed to zmert. Evaluation using BLEU works. 
Zmert uses different order of features than Moses. It is necessary to reorder
them when passing to Zmert in nbest-lists.
Previous versions used wrong copy of moses.ini, which linked to phrase tables
that were already filtered for tuning. Thus, phrase tables for evaluation were
missing a lot of phrases and the reported scores were too low.


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3092 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:42:20 +00:00
bojar
5a1c308673 Setting number of zmert iterations back to unlimited in zmert config. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3091 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:41:55 +00:00
bojar
671e7e2e4f Default size of nbest-list in zmert-moses.pl set back to 100. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3090 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:41:31 +00:00
bojar
1ce9a15483 Bug fixes im zmert-moses.pl and zmert.jar. Zmert works. 
Now it is possible to lauch Zmert with SemPOS metric.
It is possible to select a smaller model for McD parser by uncommenting line
with pdt20_train_autTag_golden_latin2_pruned_0.10.model in file zmert.tmt-scen.
Sentences are then analyzed faster (if you use TectoMT to get SemPOS tags).


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3089 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:41:11 +00:00
bojar
1b28cde2e8 small fixes of paths 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3088 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:40:29 +00:00
bojar
d47b82369f some small code modifications in zmert-moses.pl 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3087 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:39:47 +00:00
bojar
d34af9b769 updated zmert-moses.pl - zmert with sempos still not running 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3086 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:39:31 +00:00
bojar
49bb6b8882 Updated version of zmert training - still not finished SemPOS factor loading via TMT 
zmert-moses.pl - launches zmert training
zmert-decoder.pl - an is launched by zmert after each training iteration to compute scores
                   with updated lambdas
zmert.tmt-scen - TMT scenario to extract SemPOS factor for sentence without any factors


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3085 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:39:14 +00:00
bojar
90fecff3aa New mert training (Zmert) for Moses\n\nZmert jar includes SemPOS metric extension. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3084 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 22:38:53 +00:00
pjwilliams
b3f6e211fd Fix mistakes in previous commit (oh, and revert to my own svn username 
to prevent my shoddy check-ins from further sullying Hieu's good name).


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3083 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 14:02:58 +00:00
hieuhoang1972
c6d20e1f9f Update the training scripts to support the new format parameter for 
'ttable-file'


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3082 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-09 11:37:43 +00:00
sarst
7d3a22c8a7 Bugfixes for the new lexical reordering. Running without a reordering model, and with the reordering flag specyfying distance now works. Formatting of the extract.o file is now correct with 'f'-reordering models. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3054 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-06 12:08:04 +00:00
bhaddow
bd9f392875 Fix for training with non-lexicalised reordering 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3016 1f5c12ca-751b-0410-a591-d2e778427230
 2010-04-01 14:05:58 +00:00
sarst
943275e331 Set the default debug mode to 0 in train-factored-phrase-model.perl (which was the case before merging the hierarchical reordering branch to trunk) 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3009 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-30 12:09:19 +00:00
bhaddow
9573147e36 Fix --reordering-table option 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@3004 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-26 09:32:15 +00:00
bhaddow
8060024cc1 Can now specify reordering table when executing step 9 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2997 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-23 18:55:48 +00:00
bhaddow
a9920a68e1 remove extra . from ini 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2991 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-20 21:18:55 +00:00
bhaddow
4b6d3dddd2 fix error in merge 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2990 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-19 18:35:48 +00:00
bhaddow
795224736b Merge revisions 2670-2988 from track. Passes all regression except lexicalised 
reordering


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2989 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-19 17:52:51 +00:00
bgottesman
872d50dd9b Fix a line that extracts the source FACTORLIST from a FACTORMAP (in the terms of the Moses manual) so that it succeeds even if the target FACTORLIST contains commas, i.e. contains multiple factors 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2965 1f5c12ca-751b-0410-a591-d2e778427230
 2010-03-10 18:03:53 +00:00
bojar
d7335c0de1 avoid reducing corpus file if it already contains the needed factors 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2945 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-24 16:14:07 +00:00
sarst
435539d4ed Added an option to choose exactly which steps to run, not only the first and the last (since we sometimes need to do step 5 and 7 to retrain lexical reordering) 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2917 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-19 18:03:44 +00:00
sarst
b95cc2f556 Added the check from word-based models of the alignment points in the adjacent corners, to the more complex models. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2916 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-19 15:15:24 +00:00
sarst
bdfe47f99b added option to skip phrase-scoring, when we want to train new reordering models on old trainings 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2914 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-18 21:16:00 +00:00
sarst
03e24c9856 changed the maximum model config, to always print all models 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2912 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-18 13:45:15 +00:00
sarst
a86470c974 lexical reordering models: it is now possible to add the maximal reordering orientations to the extract file, and the collapsing information is no longer part of the filename. Also removed some old variables, that are not used anymore. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2911 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-18 13:07:32 +00:00
hieuhoang1972
ad3b032ebf roll back 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2910 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-18 11:00:40 +00:00
chardmeier
b10d8eb469 Added lex reo scorer to released-files and removed extra ||| from reordering table. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2894 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-12 16:23:32 +00:00
sarst
c65945b531 Cleaned up lescial reordering scoring, and sent vectors as references instead of copying them. Fixed bugs in extract: it used to choose the wrong orientation at end of sentences, and the hierarchical model typ is no longer dependent on the phrase-based model type. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2892 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-12 13:46:33 +00:00
sarst
e812b2a8e3 Bugfixes: printing the correct phrases in the table, and fixed misspelling of monotonicity. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2885 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-11 09:09:44 +00:00
sarst
92368ba490 Rewrote the lexical reordering model scoring in C++. Adapted train-factored-phrase-model.perl to that change. Minor fixes in other places, for compatibility 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2884 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-10 17:19:06 +00:00
mphi
05e21dc5e2 fixed compatibility of the --final-alignment-model 2 switch and mgiza (ibm2 is done in a single thread, hence no *.part* files) 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2859 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-04 16:12:57 +00:00
bojar
a9c2069720 switching to bash, avoiding csh's >& redirection (not accepted by dash) 
making --robust take the maximum number of restarts


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2848 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-03 14:23:53 +00:00
bojar
011840bf65 adding --force-factored-filenames to avoid problems with eager removal of '0-0' 
Conflicts:

	scripts/training/train-factored-phrase-model.perl


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2847 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-03 14:23:35 +00:00
pasmargo
275c06d9e7 Kneser-Ney and modified Kneser-Ney smoothing implementation. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2837 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-02 18:14:01 +00:00
pasmargo
63bdf3b602 reverted to previous version due to a couple of mistakes 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2836 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-02 16:51:14 +00:00
pasmargo
71b15ae0d4 Kneser-Ney and modified Kneser-Ney smoothing implementation. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2835 1f5c12ca-751b-0410-a591-d2e778427230
 2010-02-02 16:28:56 +00:00
naditomeh
242d6c6ddd word-based, phrase-based and hierarchical reordering is implemented in the training 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2823 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-31 23:56:45 +00:00
sanmarf
c37f2dc14e 4th MT marathon - lexical decomposition 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2805 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-30 10:11:35 +00:00
sarst
0785c09d03 bugfix for fatored-training config 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2804 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-30 09:34:55 +00:00
sarst
f2a5678541 added new file hierarchical.h 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2803 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-30 09:16:27 +00:00
sarst
6a13b8d186 bugfixes in train-factored-phrase-model*.perl 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2801 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-30 00:09:45 +00:00
sarst
bf70dd4767 subimtted working scripts for hierarchical training (msd) 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2796 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-29 22:38:18 +00:00
naditomeh
ad3b0760b2 adding extract.cpp 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2770 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-29 13:59:04 +00:00
naditomeh
03de8a99d8 adding extract.cpp 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2769 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-29 13:41:34 +00:00
sarst
4eec020d5b bugfixes to train-factored-phrase-model.perl 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2764 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-29 12:11:10 +00:00
sarst
a9ef19edf0 updated train-factored-phrase-model.perl to work with the new hierarchical reordering framework 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@2759 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-29 11:29:39 +00:00
hieuhoang1972
53e54def0b indent 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2723 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-28 17:17:19 +00:00
hieuhoang1972
f5ebdbcec8 xcode 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2721 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-28 16:43:36 +00:00
chardmeier
6317633148 Added memscore phrase scorer. 
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2653 1f5c12ca-751b-0410-a591-d2e778427230
 2010-01-18 14:52:34 +00:00