Changes to chart decoder cube pruning: create one cube per dotted rule

instead of one per translation and do 'non-lazy' scoring, i.e. fully
score the corner and neighbor hypotheses inside the rule cube instead
of waiting until an item is popped.  The old behaviour -- faster but
with more search errors -- is available via the
cube-pruning-lazy-scoring option.

git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@4039 1f5c12ca-751b-0410-a591-d2e778427230

1 files changed, 1 insertions, 1 deletions

diff --git a/OnDiskPt/src/TargetPhraseCollection.cpp b/OnDiskPt/src/TargetPhraseCollection.cpp
index c6364cafe..910af9ea9 100644
--- a/OnDiskPt/src/TargetPhraseCollection.cpp
+++ b/OnDiskPt/src/TargetPhraseCollection.cpp
@@ -144,7 +144,7 @@ Moses::TargetPhraseCollection *TargetPhraseCollection::ConvertToMoses(const std:
     ret->Add(mosesPhrase);
   }
 
-  ret->Prune(true, phraseDict.GetTableLimit());
+  ret->Sort(true, phraseDict.GetTableLimit());
 
   return ret;