Collinizer

src/edu/stanford/nlp/parser/lexparser/AbstractCollinizer.java

@@ -0,0 +1,14 @@
+package edu.stanford.nlp.parser.lexparser;
+
+import edu.stanford.nlp.trees.Tree;
+
+/**
+ * Interface for the Collinizers
+ *<br>
+ * TODO: pass in both the guess and the gold
+ *
+ * @author John Bauer
+ */
+public interface AbstractCollinizer  {
+  Tree transformTree(Tree guess, Tree gold);
+}

src/edu/stanford/nlp/parser/lexparser/AbstractTreebankParserParams.java

@@ -336,7 +336,7 @@ public double[] MLEDependencyGrammarSmoothingParams() {
    * tree. Should strip punctuation and maybe do some other things.
    */
   @Override
-  public abstract TreeTransformer collinizer();
+  public abstract AbstractCollinizer collinizer();
 
   /**
    * the tree transformer used to produce trees for evaluation.  Will
@@ -346,7 +346,7 @@ public double[] MLEDependencyGrammarSmoothingParams() {
    * off. (finish this doc!)
    */
   @Override
-  public abstract TreeTransformer collinizerEvalb();
+  public abstract AbstractCollinizer collinizerEvalb();
 
 
   /**

src/edu/stanford/nlp/parser/lexparser/ArabicTreebankParserParams.java

@@ -207,15 +207,15 @@ public TreeTransformer subcategoryStripper() {
    * The collinizer eliminates punctuation
    */
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(tlp, !collinizerRetainsPunctuation, false);
   }
 
   /**
    * Stand-in collinizer does nothing to the tree.
    */
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return collinizer();
   }
 

src/edu/stanford/nlp/parser/lexparser/ChineseCharacterBasedLexiconTraining.java

@@ -282,7 +282,7 @@ public static void main(String[] args) throws IOException {
       FileFilter testFilt = new NumberRangesFileFilter(testArgs[1], false);
       testTreebank.loadPath(new File(testArgs[0]), testFilt);
       TreeTransformer subcategoryStripper = op.tlpParams.subcategoryStripper();
-      TreeTransformer collinizer = ctpp.collinizer();
+      AbstractCollinizer collinizer = ctpp.collinizer();
 
       WordCatEquivalenceClasser eqclass = new WordCatEquivalenceClasser();
       WordCatEqualityChecker eqcheck = new WordCatEqualityChecker();
@@ -371,8 +371,8 @@ public static void main(String[] args) throws IOException {
           System.out.println("\nScores:");
           basicEval.displayLast();
 
-          Tree collinsTree = collinizer.transformTree(tree);
-          Tree collinsGold = collinizer.transformTree(gold);
+          Tree collinsTree = collinizer.transformTree(tree, gold);
+          Tree collinsGold = collinizer.transformTree(gold, gold);
           ourBrackets = proc.allBrackets(collinsTree);
           goldBrackets = proc.allBrackets(collinsGold);
           if (goodPOS) {

src/edu/stanford/nlp/parser/lexparser/ChineseTreebankParserParams.java

@@ -178,15 +178,15 @@ public MemoryTreebank memoryTreebank() {
    * Returns a ChineseCollinizer
    */
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new ChineseCollinizer(ctlp);
   }
 
   /**
    * Returns a ChineseCollinizer that doesn't delete punctuation
    */
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return new ChineseCollinizer(ctlp, false);
   }
 

src/edu/stanford/nlp/parser/lexparser/EnglishTreebankParserParams.java

@@ -221,12 +221,12 @@ public MemoryTreebank testMemoryTreebank() {
    * be applied both to the parser output and the gold tree.
    */
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(tlp, true, englishTrain.splitBaseNP == 2, englishTrain.collapseWhCategories);
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return new TreeCollinizer(tlp, true, englishTrain.splitBaseNP == 2, englishTrain.collapseWhCategories);
   }
 

src/edu/stanford/nlp/parser/lexparser/FactoredParser.java

@@ -461,8 +461,8 @@ public static void main(String[] args) {
         depDE.evaluate(tree3, binaryTree, pw);
         depTE.evaluate(tree3db, tree, pw);
       }
-      TreeTransformer tc = op.tlpParams.collinizer();
-      TreeTransformer tcEvalb = op.tlpParams.collinizerEvalb();
+      AbstractCollinizer tc = op.tlpParams.collinizer();
+      AbstractCollinizer tcEvalb = op.tlpParams.collinizerEvalb();
       if (op.doPCFG) {
         // System.out.println("XXXX Best PCFG was: ");
         // tree2.pennPrint();
@@ -471,8 +471,8 @@ public static void main(String[] args) {
         //System.out.println("True Best Parse:");
         //tree.pennPrint();
         //tc.transformTree(tree).pennPrint();
-        pcfgPE.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
-        pcfgCB.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
+        pcfgPE.evaluate(tc.transformTree(tree2, tree2), tc.transformTree(tree, tree), pw);
+        pcfgCB.evaluate(tc.transformTree(tree2, tree2), tc.transformTree(tree, tree), pw);
         Tree tree4b = null;
         if (op.doDep) {
           comboDE.evaluate((bothPassed ? tree4 : tree3), binaryTree, pw);
@@ -483,15 +483,15 @@ public static void main(String[] args) {
             tree4 = np.prune(tree4);
           }
           //tree4.pennPrint();
-          comboPE.evaluate(tc.transformTree(tree4), tc.transformTree(tree), pw);
+          comboPE.evaluate(tc.transformTree(tree4, tree4), tc.transformTree(tree, tree), pw);
         }
         //pcfgTE.evaluate(tree2, tree);
-        pcfgTE.evaluate(tcEvalb.transformTree(tree2), tcEvalb.transformTree(tree), pw);
-        pcfgTEnoPunct.evaluate(tc.transformTree(tree2), tc.transformTree(tree), pw);
+        pcfgTE.evaluate(tcEvalb.transformTree(tree2, tree2), tcEvalb.transformTree(tree, tree), pw);
+        pcfgTEnoPunct.evaluate(tc.transformTree(tree2, tree2), tc.transformTree(tree, tree), pw);
 
         if (op.doDep) {
-          comboTE.evaluate(tcEvalb.transformTree(tree4), tcEvalb.transformTree(tree), pw);
-          comboTEnoPunct.evaluate(tc.transformTree(tree4), tc.transformTree(tree), pw);
+          comboTE.evaluate(tcEvalb.transformTree(tree4, tree4), tcEvalb.transformTree(tree, tree), pw);
+          comboTEnoPunct.evaluate(tc.transformTree(tree4, tree4), tc.transformTree(tree, tree), pw);
         }
         System.out.println("PCFG only: " + parser.scoreBinarizedTree(tree2b, 0));
 
@@ -515,11 +515,11 @@ public static void main(String[] args) {
 
       if (op.testOptions.evalb) {
         if (op.doPCFG && op.doDep) {
-          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree4));
+          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree, tree), tcEvalb.transformTree(tree4, tree4));
         } else if (op.doPCFG) {
-          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree2));
+          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree, tree), tcEvalb.transformTree(tree2, tree2));
         } else if (op.doDep) {
-          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree), tcEvalb.transformTree(tree3db));
+          EvalbFormatWriter.writeEVALBline(tcEvalb.transformTree(tree, tree), tcEvalb.transformTree(tree3db, tree3db));
         }
       }
     } // end for each tree in test treebank

src/edu/stanford/nlp/parser/lexparser/FrenchTreebankParserParams.java

@@ -549,12 +549,12 @@ public String[] sisterSplitters() {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(treebankLanguagePack());
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return new TreeCollinizer(treebankLanguagePack(),collinizerRetainsPunctuation,false);
   }
 

src/edu/stanford/nlp/parser/lexparser/GenericTreebankParserParams.java

@@ -24,13 +24,13 @@ protected GenericTreebankParserParams(TreebankLanguagePack tlp) {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     // TODO Auto-generated method stub
     return null;
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     // TODO Auto-generated method stub
     return null;
   }

src/edu/stanford/nlp/parser/lexparser/HebrewTreebankParserParams.java

@@ -47,15 +47,15 @@ protected HebrewTreebankParserParams(TreebankLanguagePack tlp) {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(tlp, true, false);
   }
 
   /**
    * Stand-in collinizer does nothing to the tree.
    */
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return collinizer();
   }
 

src/edu/stanford/nlp/parser/lexparser/HungarianTreebankParserParams.java

@@ -53,12 +53,12 @@ public HeadFinder typedDependencyHeadFinder() {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(tlp, true, false, 0);
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return collinizer();
   }
 

src/edu/stanford/nlp/parser/lexparser/ItalianTreebankParserParams.java

@@ -53,12 +53,12 @@ public HeadFinder typedDependencyHeadFinder() {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(tlp, true, false, 0);
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return collinizer();
   }
 

src/edu/stanford/nlp/parser/lexparser/NegraPennCollinizer.java

@@ -2,17 +2,18 @@
 import edu.stanford.nlp.util.logging.Redwood;
 
 import java.util.ArrayList;
+import java.util.Iterator;
 import java.util.List;
 
 import edu.stanford.nlp.ling.Label;
 import edu.stanford.nlp.ling.StringLabel;
 import edu.stanford.nlp.trees.LabeledScoredTreeFactory;
 import edu.stanford.nlp.trees.Tree;
+import edu.stanford.nlp.trees.Trees;
 import edu.stanford.nlp.trees.TreeFactory;
-import edu.stanford.nlp.trees.TreeTransformer;
 
 
-public class NegraPennCollinizer implements TreeTransformer {
+public class NegraPennCollinizer implements AbstractCollinizer {
 
   /** A logger for this class */
   Redwood.RedwoodChannels log = Redwood.channels(NegraPennCollinizer.class);
@@ -31,37 +32,47 @@ public NegraPennCollinizer(TreebankLangParserParams tlpp, boolean deletePunct) {
 
   protected TreeFactory tf = new LabeledScoredTreeFactory();
 
-  public Tree transformTree(Tree tree) {
-    Label l = tree.label();
-    if (tree.isLeaf()) {
+  public Tree transformTree(Tree guess, Tree gold) {
+    if (guess == null || gold == null) return null;
+    if (guess.yield().size() != gold.yield().size()) {
+      return null;
+    }
+
+    return transformTree(guess, Trees.preTerminals(gold).iterator());
+  }
+
+  private Tree transformTree(Tree guess, Iterator<Tree> goldPreterminals) {
+    Label l = guess.label();
+    if (guess.isLeaf()) {
       return tf.newLeaf(l);
     }
     String s = l.value();
     s = tlpp.treebankLanguagePack().basicCategory(s);
-    if (deletePunct) {
-      // this is broken as it's not the right thing to do when there
-      // is any tag ambiguity -- and there is for ' (POS/'').  Sentences
-      // can then have more or less words.  It's also unnecessary for EVALB,
-      // since it ignores punctuation anyway
-      if (tree.isPreTerminal() && tlpp.treebankLanguagePack().isEvalBIgnoredPunctuationTag(s)) {
+    if (deletePunct && guess.isPreTerminal()) {
+      // Eliminate unwanted (in terms of evaluation) punctuation
+      // by comparing the gold punctuation, not the guess tree
+      // This way, retagging does not change the results
+      Tree goldPT = goldPreterminals.next();
+      String goldTag = tlpp.treebankLanguagePack().basicCategory(goldPT.value());
+      if (tlpp.treebankLanguagePack().isEvalBIgnoredPunctuationTag(goldTag)) {
         return null;
       }
     }
     // TEMPORARY: eliminate the TOPP constituent
-    if (tree.children()[0].label().value().equals("TOPP")) {
+    if (guess.children()[0].label().value().equals("TOPP")) {
       log.info("Found a TOPP");
-      tree.setChildren(tree.children()[0].children());
+      guess.setChildren(guess.children()[0].children());
     }
 
     // Negra has lots of non-unary roots; delete unary roots
-    if (tlpp.treebankLanguagePack().isStartSymbol(s) && tree.numChildren() == 1) {
+    if (tlpp.treebankLanguagePack().isStartSymbol(s) && guess.numChildren() == 1) {
       // NB: This deletes the boundary symbol, which is in the tree!
-      return transformTree(tree.getChild(0));
+      return transformTree(guess.getChild(0), goldPreterminals);
     }
     List<Tree> children = new ArrayList<>();
-    for (int cNum = 0, numC = tree.numChildren(); cNum < numC; cNum++) {
-      Tree child = tree.getChild(cNum);
-      Tree newChild = transformTree(child);
+    for (int cNum = 0, numC = guess.numChildren(); cNum < numC; cNum++) {
+      Tree child = guess.getChild(cNum);
+      Tree newChild = transformTree(child, goldPreterminals);
       if (newChild != null) {
         children.add(newChild);
       }

src/edu/stanford/nlp/parser/lexparser/NegraPennTreebankParserParams.java

@@ -125,15 +125,15 @@ public DiskTreebank diskTreebank() {
    * returns a NegraPennCollinizer
    */
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new NegraPennCollinizer(this);
   }
 
   /**
    * returns a NegraPennCollinizer
    */
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return new NegraPennCollinizer(this, false);
   }
 

src/edu/stanford/nlp/parser/lexparser/SpanishTreebankParserParams.java

@@ -262,12 +262,12 @@ public String[] sisterSplitters() {
   }
 
   @Override
-  public TreeTransformer collinizer() {
+  public AbstractCollinizer collinizer() {
     return new TreeCollinizer(treebankLanguagePack());
   }
 
   @Override
-  public TreeTransformer collinizerEvalb() {
+  public AbstractCollinizer collinizerEvalb() {
     return new TreeCollinizer(treebankLanguagePack());
   }