single word heuristic for phrase extraction,

and minor modification of SentenceAlignmentWithSyntax constructor

6 files changed, 30 insertions, 19 deletions

diff --git a/phrase-extract/PhraseExtractionOptions.h b/phrase-extract/PhraseExtractionOptions.h
index cf2e4b365..38e570b79 100644
--- a/phrase-extract/PhraseExtractionOptions.h
+++ b/phrase-extract/PhraseExtractionOptions.h
@@ -51,6 +51,7 @@ private:
   bool gzOutput;
   std::string instanceWeightsFile; //weights for each sentence
   bool flexScoreFlag;
+  bool singleWordHeuristicFlag;
 
 public:
   std::vector<std::string> placeholders;
@@ -73,6 +74,7 @@ public:
     onlyOutputSpanInfo(false),
     gzOutput(false),
     flexScoreFlag(false),
+    singleWordHeuristicFlag(false),
     debug(false) {
   }
 
@@ -119,6 +121,9 @@ public:
   void initFlexScoreFlag(const bool initflexScoreFlag) {
     flexScoreFlag=initflexScoreFlag;
   }
+  void initSingleWordHeuristicFlag(const bool initSingleWordHeuristicFlag) {
+    singleWordHeuristicFlag = initSingleWordHeuristicFlag;
+  }
 
   // functions for getting values
   bool isAllModelsOutputFlag() const {
@@ -163,6 +168,9 @@ public:
   bool isFlexScoreFlag() const {
     return flexScoreFlag;
   }
+  bool isSingleWordHeuristicFlag() const {
+    return singleWordHeuristicFlag;
+  }
 };
 
 }
diff --git a/phrase-extract/RuleExtractionOptions.h b/phrase-extract/RuleExtractionOptions.h
index 43f7b205b..7dfb90177 100644
--- a/phrase-extract/RuleExtractionOptions.h
+++ b/phrase-extract/RuleExtractionOptions.h
@@ -18,8 +18,6 @@
  ***********************************************************************/
 
 #pragma once
-#ifndef RULEEXTRACTIONOPTIONS_H_INCLUDED_
-#define RULEEXTRACTIONOPTIONS_H_INCLUDED_
 
 namespace MosesTraining
 {
@@ -95,4 +93,3 @@ public:
 
 }
 
-#endif
diff --git a/phrase-extract/SentenceAlignmentWithSyntax.cpp b/phrase-extract/SentenceAlignmentWithSyntax.cpp
index 4fd2355ae..89361a45d 100644
--- a/phrase-extract/SentenceAlignmentWithSyntax.cpp
+++ b/phrase-extract/SentenceAlignmentWithSyntax.cpp
@@ -35,7 +35,7 @@ namespace MosesTraining
 
 bool SentenceAlignmentWithSyntax::processTargetSentence(const char * targetString, int sentenceID, bool boundaryRules)
 {
-  if (!m_options.targetSyntax) {
+  if (!m_targetSyntax) {
     return SentenceAlignment::processTargetSentence(targetString, sentenceID, boundaryRules);
   }
 
@@ -56,7 +56,7 @@ bool SentenceAlignmentWithSyntax::processTargetSentence(const char * targetStrin
 
 bool SentenceAlignmentWithSyntax::processSourceSentence(const char * sourceString, int sentenceID, bool boundaryRules)
 {
-  if (!m_options.sourceSyntax) {
+  if (!m_sourceSyntax) {
     return SentenceAlignment::processSourceSentence(sourceString, sentenceID, boundaryRules);
   }
 
diff --git a/phrase-extract/SentenceAlignmentWithSyntax.h b/phrase-extract/SentenceAlignmentWithSyntax.h
index 604b6d0e2..be7ee52d3 100644
--- a/phrase-extract/SentenceAlignmentWithSyntax.h
+++ b/phrase-extract/SentenceAlignmentWithSyntax.h
@@ -18,8 +18,6 @@
  ***********************************************************************/
 
 #pragma once
-#ifndef SENTENCEALIGNMENTWITHSYNTAX_H_INCLUDED_
-#define SENTENCEALIGNMENTWITHSYNTAX_H_INCLUDED_
 
 #include <map>
 #include <set>
@@ -42,18 +40,20 @@ public:
   std::set<std::string> & m_sourceLabelCollection;
   std::map<std::string, int> & m_targetTopLabelCollection;
   std::map<std::string, int> & m_sourceTopLabelCollection;
-  const RuleExtractionOptions & m_options;
+  const bool m_targetSyntax, m_sourceSyntax;
 
   SentenceAlignmentWithSyntax(std::set<std::string> & tgtLabelColl,
                               std::set<std::string> & srcLabelColl,
                               std::map<std::string,int> & tgtTopLabelColl,
                               std::map<std::string,int> & srcTopLabelColl,
-                              const RuleExtractionOptions & options)
+                              bool targetSyntax,
+                              bool sourceSyntax)
     : m_targetLabelCollection(tgtLabelColl)
     , m_sourceLabelCollection(srcLabelColl)
     , m_targetTopLabelCollection(tgtTopLabelColl)
     , m_sourceTopLabelCollection(srcTopLabelColl)
-    , m_options(options) {
+    , m_targetSyntax(targetSyntax)
+    , m_sourceSyntax(sourceSyntax) {
   }
 
   virtual ~SentenceAlignmentWithSyntax() {}
@@ -67,4 +67,3 @@ public:
 
 }
 
-#endif
diff --git a/phrase-extract/extract-main.cpp b/phrase-extract/extract-main.cpp
index c23fda489..ef989fac6 100644
--- a/phrase-extract/extract-main.cpp
+++ b/phrase-extract/extract-main.cpp
@@ -155,6 +155,8 @@ int main(int argc, char* argv[])
       options.initOrientationFlag(true);
     } else if (strcmp(argv[i],"--FlexibilityScore") == 0) {
       options.initFlexScoreFlag(true);
+    } else if (strcmp(argv[i],"--SingleWordHeuristic") == 0) {
+      options.initSingleWordHeuristicFlag(true);
     } else if (strcmp(argv[i],"--NoTTable") == 0) {
       options.initTranslationFlag(false);
     } else if (strcmp(argv[i], "--IncludeSentenceId") == 0) {
@@ -413,18 +415,22 @@ void ExtractTask::extract(SentenceAlignment &sentence)
           }
 
         // cout << "doing if for ( " << minF << "-" << maxF << ", " << startE << "," << endE << ")\n";
-        if (!out_of_bounds) {
+        if (!out_of_bounds || 
+            ( m_options.isSingleWordHeuristicFlag() && (endE==startE) && (minF==maxF) )) // extraction of single word phrases even if inconsistent wrt. word alignment
+        {
           // start point of source phrase may retreat over unaligned
           for(int startF=minF;
-              (startF>=0 &&
-               (relaxLimit || startF>maxF-m_options.maxPhraseLength) && // within length limit
-               (startF==minF || sentence.alignedCountS[startF]==0)); // unaligned
+              ((startF>=0 &&
+                (relaxLimit || startF>maxF-m_options.maxPhraseLength) && // within length limit
+                (startF==minF || sentence.alignedCountS[startF]==0)) && // unaligned
+               (!out_of_bounds || (startF==minF))); // if out of bounds, but single word heuristic: don't retreat over unaligned
               startF--)
             // end point of source phrase may advance over unaligned
             for(int endF=maxF;
-                (endF<countF &&
-                 (relaxLimit || endF<startF+m_options.maxPhraseLength) && // within length limit
-                 (endF==maxF || sentence.alignedCountS[endF]==0)); // unaligned
+                ((endF<countF &&
+                  (relaxLimit || endF<startF+m_options.maxPhraseLength) && // within length limit
+                  (endF==maxF || sentence.alignedCountS[endF]==0)) && // unaligned
+                 (!out_of_bounds || (endF==maxF))); // if out of bounds, but single word heuristic: don't advance over unaligned
                 endF++) { // at this point we have extracted a phrase
               if(buildExtraStructure) { // phrase || hier
                 if(endE-startE < m_options.maxPhraseLength && endF-startF < m_options.maxPhraseLength) { // within limit
diff --git a/phrase-extract/extract-rules-main.cpp b/phrase-extract/extract-rules-main.cpp
index 125383a7e..711918656 100644
--- a/phrase-extract/extract-rules-main.cpp
+++ b/phrase-extract/extract-rules-main.cpp
@@ -347,7 +347,8 @@ int main(int argc, char* argv[])
 
     SentenceAlignmentWithSyntax sentence
     (targetLabelCollection, sourceLabelCollection,
-     targetTopLabelCollection, sourceTopLabelCollection, options);
+     targetTopLabelCollection, sourceTopLabelCollection, 
+     options.targetSyntax, options.sourceSyntax);
     //az: output src, tgt, and alingment line
     if (options.onlyOutputSpanInfo) {
       cout << "LOG: SRC: " << sourceString << endl;