Commit Graph

53 Commits

Author SHA1 Message Date
bhaddow
407dd68aec Update lexicalised reordering test truths to take account of new, improved scores.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/branches/hierarchical-reo@3001 1f5c12ca-751b-0410-a591-d2e778427230
2010-03-25 11:04:17 +00:00
hieuhoang1972
cbb329b754 minor
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2955 1f5c12ca-751b-0410-a591-d2e778427230
2010-03-07 19:39:07 +00:00
chardmeier
917737d1eb Fixed regression test for lattice distortion because implementation is now right and regression test isn't.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2652 1f5c12ca-751b-0410-a591-d2e778427230
2010-01-18 11:13:39 +00:00
phkoehn
8833098925 generalized n-best list reporting for feature functions, added experimental version of global lexical model
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2343 1f5c12ca-751b-0410-a591-d2e778427230
2009-05-26 19:30:35 +00:00
redpony
558c723c7c fix for one round of problems- this moses will work with g++ > 4.3
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2236 1f5c12ca-751b-0410-a591-d2e778427230
2009-03-12 16:52:27 +00:00
redpony
63effe85b5 Feature function overhaul. Each feature function is computed in one of three ways:
1) Stateless feature functions from the phrase table/generation table: these are computed when
   the TranslationOption is created.  They become part of the ScoreBreakdown object contained in
   the TranslationOption and are added to the feature value vector when a hypothesis is extended.
2) Stateless feature functions that are computed during state exploration. Currently, only
   WordPenalty falls into this category, but these functions implement a method Evaluate which
   do does not receive a Hypothesis or any contextual information.
3) Stateful feature functions: these features receive the arc information (translation option),
   compute some value and then return some context information.  The context information created
   by a particular feature function is passed back to it as the previous context when a hypothesis
   originating at the node where the previous edge terminates is created.  States in the search
   space may be recombined if the context information is identical.  The context information must
   be stored in an object implementing the FFState interface.

TODO:
1) the command line interface / MERT interface needs to go to named parameters that are otherwise opaque
2) StatefulFeatureFunction's Evaluate method should just take a TranslationOption and a context object.  It is not good that it takes a hypothesis, because then people may be tempted to access information about the "previous" hypothesis without "declaring" this dependency.
3) Future cost estimates should be handled using feature functions.  All stateful feature functions need some kind of future cost estimate.
4) Philipp's poor-man's cube pruning is broken.



git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@2087 1f5c12ca-751b-0410-a591-d2e778427230
2009-02-06 15:43:06 +00:00
redpony
02067b2be0 fix some nasty edge cases in lattice decoding that arise when decoding lattices with complex topologies. also fix lattice regression tests. work done jointly with j schroeder.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1853 1f5c12ca-751b-0410-a591-d2e778427230
2008-06-17 17:32:19 +00:00
hieuhoang1972
4f84e3ccfe don't run 2 tests which don't pass
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1850 1f5c12ca-751b-0410-a591-d2e778427230
2008-06-11 13:42:23 +00:00
redpony
aed1758623 fix bug in word lattice decoding that prevented some viable hypotheses from being considered.
also fix XML-markup test failures.


git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1563 1f5c12ca-751b-0410-a591-d2e778427230
2008-02-19 22:03:05 +00:00
redpony
9521e78fbf add functionalty to support adding direct translation (ie, passthrouh) even when there are other translation options available for the span.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1510 1f5c12ca-751b-0410-a591-d2e778427230
2007-11-10 23:19:41 +00:00
redpony
180d9bac5d add support for unary XML options. refer to regression-testing/tests/xml-markup/to-translate.txt for examples.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1508 1f5c12ca-751b-0410-a591-d2e778427230
2007-11-10 04:40:18 +00:00
redpony
3b008f67a9 add support for multiple xml options for the same span, start refactoring to add support for overlapping xml options.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1507 1f5c12ca-751b-0410-a591-d2e778427230
2007-11-09 23:56:57 +00:00
redpony
7dbcef925d Implemented -r (-recover-input-path) to recover the actual path through the CN/word lattice that was used in the best translation. Also cleaned up handling of source Phrases in the context of a particular hypothesis, which may improve performance translating with lexicalized reordering models.
Removed PhraseReference since the concept was broken for CN/lattice inputs.



git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1468 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-28 16:43:33 +00:00
redpony
17884106d4 is this the last of the word lattice distortion errors? maybe. i hope so. the nbest list is at least in the right order now.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1466 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-27 23:22:58 +00:00
redpony
541a486c0c stricter checks on word lattice distortion
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1464 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-27 12:14:42 +00:00
redpony
251aa600fa fixed a subtle bug in the word lattice decoder. it now acts sane. This experience also gave me more evidence for thinking that numbering the words in a phrase is not good-- we should be numbering the spaces between them. this would fix oh so many nasty things we've got (like the fact that we represent word drops with [n,n-1] etc).
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1463 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-27 05:03:06 +00:00
redpony
81c0642e36 fix distortion limiting functionality for WordLattices. Now, the shortest path to a another node is used to determine whether the threshold has been exceeded.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1460 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-17 02:49:08 +00:00
redpony
63da638077 The primary change here is that the distortion distance in a word lattice is now defined to be the number of words on the shortest path between the previous phrase's ending span and the next phrase's starting span. Originally, the distance was defined to be the difference in node number, which could be extremely large even for words that are, in some sense, adjacent. Only the distortion scoring has been fixed for word lattices, but the distortion limit is still broken.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1458 1f5c12ca-751b-0410-a591-d2e778427230
2007-09-14 07:31:56 +00:00
redpony
d8536d15b7 add lattice test, not just CN test
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1422 1f5c12ca-751b-0410-a591-d2e778427230
2007-07-03 15:40:26 +00:00
redpony
bf756e12fb add test
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1363 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-18 17:03:38 +00:00
redpony
945c77e3f2 fix bug with unknown words/CNs/lexical reordering
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1362 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-18 16:49:51 +00:00
redpony
c80d8b8d47 Support for the decoding of arbitrary word lattices. Must be given in the form of a "plf" file, which is a little tricky. I'll add documentation at some point; for now, refer to the example plf file in the "lattice-surface" regression test.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1359 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-18 14:08:46 +00:00
hieuhoang1972
f4a02f840d changed lex reordering regression test to fit with philipp's change to staticData
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1346 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-08 22:15:14 +00:00
hieuhoang1972
d48e71cb57 cleanup regression files - consistent file extension, delete old ini files
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1344 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-06 17:07:54 +00:00
hieuhoang1972
b6945f91bc change user message for downloading data
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1343 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-06 14:06:39 +00:00
hieuhoang1972
10ee6bedf5 update regression data
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1342 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-06 13:55:45 +00:00
hieuhoang1972
fa0231aea9 add lexicalized reordering test
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1341 1f5c12ca-751b-0410-a591-d2e778427230
2007-04-06 13:25:20 +00:00
hieuhoang1972
3c07c5df4d add svn id comments to start of file
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1307 1f5c12ca-751b-0410-a591-d2e778427230
2007-03-14 22:22:36 +00:00
hieuhoang1972
5f22fb13d3 make output of decimal places consistent by not format anywhere but in Main.cpp
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1266 1f5c12ca-751b-0410-a591-d2e778427230
2007-03-04 00:43:45 +00:00
hieuhoang1972
6b611279d5 minor gcc compile error.
also, no longer use IRSTLM as a subsitutute for SRILM, and vice versa. They don't give identitcal results - avoids confusion.

git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1229 1f5c12ca-751b-0410-a591-d2e778427230
2007-02-21 20:05:27 +00:00
hieuhoang1972
b62dda41ed change unknown word processing to be closer to the way pharaoh does it - create unknown word whenever single word is not in translation table but penalise hypothesis for using it.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1228 1f5c12ca-751b-0410-a591-d2e778427230
2007-02-21 19:51:17 +00:00
hieuhoang1972
7ecb0ce66e change unknown word processing to be closer to the way pharaoh does it - create unknown word whenever single word is not in translation table but penalise hypothesis for using it.
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1227 1f5c12ca-751b-0410-a591-d2e778427230
2007-02-21 19:48:53 +00:00
hieuhoang1972
e247f1da6f fixed regression test failing. Number of features for generation models MUST be specified in ini file, no backward compatability hack
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1209 1f5c12ca-751b-0410-a591-d2e778427230
2007-02-13 19:15:34 +00:00
hieuhoang1972
afe5f3bf2e undo update to regression data version
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@1060 1f5c12ca-751b-0410-a591-d2e778427230
2006-12-13 12:41:15 +00:00
hieuhoang1972
abd4adfb0c undo accidental checkin for MosesRegressionTesting.pm
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@999 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-22 15:50:59 +00:00
hieuhoang1972
dd721e1b4f Mac xcode proj
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@997 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-22 15:19:28 +00:00
nicolabertoldi
90d9147d9e Add simple regression test for nbest
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@986 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-20 13:42:04 +00:00
nicolabertoldi
7f7e7dfbd9 Add regression test for confusion network and multiple factors
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@985 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-20 13:27:24 +00:00
hieuhoang1972
4e2f8cf550 consistent file extension
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@967 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-09 06:55:23 +00:00
hieuhoang1972
a373d97ecf consistent file extension
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@966 1f5c12ca-751b-0410-a591-d2e778427230
2006-11-08 22:44:29 +00:00
bojar
27fe632194 being safe and dying when unknown --cmdline-option supplied
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@824 1f5c12ca-751b-0410-a591-d2e778427230
2006-09-21 14:45:03 +00:00
redpony
c69cfaf33e Allow the factor delimiter, that is, the string that separates the factors in a 'word' to be specified to moses and to train-factored-phrase-model.perl. The default is still to use '|'. Multi-character delimiters are allowed (for example, '+++'). Added a regression test for multi-character delimiters.
Remove JHU dependencies on make release.  It now looks for GIZA++ and sets the BINDIR inside train-factored-phrase-model.perl at installation time (note: because of this, this script MUST BE released before it can be run now).



git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@812 1f5c12ca-751b-0410-a591-d2e778427230
2006-09-12 15:53:50 +00:00
phkoehn
1d7bf5f8c2 regression tests had to adjusted to changes in verbose handling
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@801 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-30 17:05:45 +00:00
hieuhoang1972
1e63b42c06 fixed to run from any directory
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@661 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 18:36:23 +00:00
hieuhoang1972
47ab16ba5d fixed to work with most recent moses (8/11 1429) and to use good coding style (yay perl module!)
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@660 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 18:25:01 +00:00
hieuhoang1972
80e0f752fa fixed to work with most recent moses as of now (8/11 1429) and to use good coding style (yay perl module!)
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@659 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 18:24:15 +00:00
redpony
c8d3471bbe fix tests
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@657 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 18:16:00 +00:00
eherbst
b340381eb4 put a couple common functions into a utils perl module; fixed a bit of broken format-checking
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@652 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 16:35:50 +00:00
hieuhoang1972
99de4efd3c eclipse proj
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@650 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 16:13:42 +00:00
eherbst
0afacafddf adding characters to the printing of output dir
git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@645 1f5c12ca-751b-0410-a591-d2e778427230
2006-08-11 15:00:59 +00:00