From 9a38a7dba975a0ac2d287bb5ab6161182499730f Mon Sep 17 00:00:00 2001
From: Eva Hasler <ehasler@saxnot.inf.ed.ac.uk>
Date: Fri, 17 Feb 2012 15:55:59 +0000
Subject: [PATCH] add optional bias feature to GLMU

---
 moses/src/GlobalLexicalModelUnlimited.cpp | 33 +++++++++++++++--------
 moses/src/GlobalLexicalModelUnlimited.h   |  4 ++-
 moses/src/StaticData.cpp                  | 16 ++++++-----
 3 files changed, 35 insertions(+), 18 deletions(-)

diff --git a/moses/src/GlobalLexicalModelUnlimited.cpp b/moses/src/GlobalLexicalModelUnlimited.cpp
index ce548f89c..14e9d821d 100644
--- a/moses/src/GlobalLexicalModelUnlimited.cpp
+++ b/moses/src/GlobalLexicalModelUnlimited.cpp
@@ -10,10 +10,12 @@ namespace Moses
 {
 GlobalLexicalModelUnlimited::GlobalLexicalModelUnlimited(const vector< FactorType >& inFactors,
                                                          const vector< FactorType >& outFactors,
-                                                         bool ignorePunctuation)
+                                                         bool ignorePunctuation,
+							 bool biasFeature)
 : StatelessFeatureFunction("glm",ScoreProducer::unlimited),
   m_sparseProducerWeight(1),
-  m_ignorePunctuation(ignorePunctuation)
+  m_ignorePunctuation(ignorePunctuation),
+  m_biasFeature(biasFeature)
 {
 	std::cerr << "Creating global lexical model unlimited.. ";
 
@@ -22,12 +24,12 @@ GlobalLexicalModelUnlimited::GlobalLexicalModelUnlimited(const vector< FactorTyp
 
 	// compile a list of punctuation characters
 	if (m_ignorePunctuation) {
-		cerr << "ignoring punctuation";
+		cerr << "ignoring punctuation.. ";
 		char punctuation[] = "\"'!?¿·()#_,.:;•&@‑/\\0123456789~=";
 		for (size_t i=0; i < sizeof(punctuation)-1; ++i)
 			m_punctuationHash[punctuation[i]] = 1;
 	}
-	cerr << endl;
+	cerr << "done." << endl;
 }
 
 GlobalLexicalModelUnlimited::~GlobalLexicalModelUnlimited(){}
@@ -53,13 +55,22 @@ void GlobalLexicalModelUnlimited::Evaluate(const TargetPhrase& targetPhrase, Sco
   	string targetString = targetPhrase.GetWord(targetIndex).GetString(0); // TODO: change for other factors
 
   	if (m_ignorePunctuation) {
-  		// check if first char is punctuation
-  		char firstChar = targetString.at(0);
-  		CharHash::const_iterator charIterator = m_punctuationHash.find( firstChar );
-  		if(charIterator != m_punctuationHash.end())
-  			continue;
+	  // check if first char is punctuation
+	  char firstChar = targetString.at(0);
+	  CharHash::const_iterator charIterator = m_punctuationHash.find( firstChar );
+	  if(charIterator != m_punctuationHash.end())
+	    continue;	 
   	}
 
+	if (m_biasFeature) {
+	  stringstream feature;
+	  feature << "glm_";
+	  feature << targetString;
+	  feature << "~";
+	  feature << "**BIAS**";
+	  accumulator->SparsePlusEquals(feature.str(), 1);
+	}
+
 //  	set< const Word*, WordComparer > alreadyScored; // do not score a word twice
   	StringHash alreadyScored;
   	for(size_t inputIndex = 0; inputIndex < input.GetSize(); inputIndex++ ) {
@@ -69,8 +80,8 @@ void GlobalLexicalModelUnlimited::Evaluate(const TargetPhrase& targetPhrase, Sco
   			// check if first char is punctuation
   			char firstChar = inputString.at(0);
   			CharHash::const_iterator charIterator = m_punctuationHash.find( firstChar );
-  			if(charIterator != m_punctuationHash.end())
-  				continue;
+  			if(charIterator != m_punctuationHash.end()) 
+			  continue;			
   		}
 
   		//if ( alreadyScored.find( &inputWord ) == alreadyScored.end() ) {
diff --git a/moses/src/GlobalLexicalModelUnlimited.h b/moses/src/GlobalLexicalModelUnlimited.h
index 19476c679..e867dc116 100644
--- a/moses/src/GlobalLexicalModelUnlimited.h
+++ b/moses/src/GlobalLexicalModelUnlimited.h
@@ -53,6 +53,7 @@ private:
 
   CharHash m_punctuationHash;
   bool m_ignorePunctuation;
+  bool m_biasFeature;
 
   std::vector< FactorType > m_inputFactors;
   std::vector< FactorType > m_outputFactors;
@@ -65,7 +66,8 @@ private:
 public:
   GlobalLexicalModelUnlimited(const std::vector< FactorType >& inFactors,
 	                            const std::vector< FactorType >& outFactors,
-	                            bool ignorePunctuation);
+			            bool ignorePunctuation,
+			            bool biasFeature);
 
   virtual ~GlobalLexicalModelUnlimited();
 
diff --git a/moses/src/StaticData.cpp b/moses/src/StaticData.cpp
index af5861b22..6f6636d2f 100644
--- a/moses/src/StaticData.cpp
+++ b/moses/src/StaticData.cpp
@@ -973,13 +973,17 @@ bool StaticData::LoadGlobalLexicalModelUnlimited()
 
   for (size_t i = 0; i < weight.size(); i++ ) {
     bool ignorePunctuation = false;
+    bool biasFeature = false;
     vector< string > factors;
-    vector< string > factors_punctuation = Tokenize(modelSpec[i]," ");
+    vector< string > factors_punct_bias = Tokenize(modelSpec[i]," ");
 
-    // read optional punctuation specification
-    if (factors_punctuation.size() > 0) {
-    	factors = Tokenize(factors_punctuation[0],"-");
-    	ignorePunctuation = Scan<int>(factors_punctuation[1]);
+    // read optional punctuation and bias specifications
+    if (factors_punct_bias.size() > 0) {
+    	factors = Tokenize(factors_punct_bias[0],"-");
+	if (factors_punct_bias.size() >= 2)
+	  ignorePunctuation = Scan<int>(factors_punct_bias[1]);
+	if (factors_punct_bias.size() >= 3)
+	  biasFeature = Scan<int>(factors_punct_bias[2]);
     }
     else
     	factors = Tokenize(modelSpec[i],"-");
@@ -990,7 +994,7 @@ bool StaticData::LoadGlobalLexicalModelUnlimited()
     }
     const vector<FactorType> inputFactors = Tokenize<FactorType>(factors[0],",");
     const vector<FactorType> outputFactors = Tokenize<FactorType>(factors[1],",");
-    m_globalLexicalModelsUnlimited.push_back(new GlobalLexicalModelUnlimited(inputFactors, outputFactors, ignorePunctuation));
+    m_globalLexicalModelsUnlimited.push_back(new GlobalLexicalModelUnlimited(inputFactors, outputFactors, ignorePunctuation, biasFeature));
     m_globalLexicalModelsUnlimited[i]->SetSparseProducerWeight(weight[i]);
   }
   return true;