Ejemplo n.º 1
0
 Tree prune(Tree tree, int start) {
   if (tree.isLeaf() || tree.isPreTerminal()) {
     return tree;
   }
   // check each node's children for deletion
   List<Tree> children = helper(tree.getChildrenAsList(), start);
   children = prune(children, tree.label(), start, start + tree.yield().size());
   return tree.treeFactory().newTreeNode(tree.label(), children);
 }
Ejemplo n.º 2
0
 private static <L> void appendPreTerminalYield(Tree<L> tree, List<L> yield) {
   if (tree.isPreTerminal()) {
     yield.add(tree.getLabel());
     return;
   }
   for (Tree<L> child : tree.getChildren()) {
     appendPreTerminalYield(child, yield);
   }
 }
Ejemplo n.º 3
0
 private static void preTerminals(Tree t, List<Tree> l) {
   if (t.isPreTerminal()) {
     l.add(t);
   } else {
     Tree[] kids = t.children();
     for (int j = 0, n = kids.length; j < n; j++) {
       preTerminals(kids[j], l);
     }
   }
 }
Ejemplo n.º 4
0
 private static <L> int toConstituentCollectionHelper(
     Tree<L> tree, int start, List<Constituent<L>> constituents) {
   if (tree.isLeaf() || tree.isPreTerminal()) return 1;
   int span = 0;
   for (Tree<L> child : tree.getChildren()) {
     span += toConstituentCollectionHelper(child, start + span, constituents);
   }
   constituents.add(new Constituent<L>(tree.getLabel(), start, start + span));
   return span;
 }
Ejemplo n.º 5
0
  /**
   * This is the method to call for assigning labels and node vectors to the Tree. After calling
   * this, each of the non-leaf nodes will have the node vector and the predictions of their classes
   * assigned to that subtree's node.
   */
  public void forwardPropagateTree(Tree tree) {
    FloatMatrix nodeVector;
    FloatMatrix classification;

    if (tree.isLeaf()) {
      // We do nothing for the leaves.  The preterminals will
      // calculate the classification for this word/tag.  In fact, the
      // recursion should not have gotten here (unless there are
      // degenerate trees of just one leaf)
      throw new AssertionError("We should not have reached leaves in forwardPropagate");
    } else if (tree.isPreTerminal()) {
      classification = getUnaryClassification(tree.label());
      String word = tree.children().get(0).value();
      FloatMatrix wordVector = getFeatureVector(word);
      if (wordVector == null) {
        wordVector = featureVectors.get(UNKNOWN_FEATURE);
      }

      nodeVector = activationFunction.apply(wordVector);
    } else if (tree.children().size() == 1) {
      throw new AssertionError(
          "Non-preterminal nodes of size 1 should have already been collapsed");
    } else if (tree.children().size() == 2) {
      Tree left = tree.firstChild(), right = tree.lastChild();
      forwardPropagateTree(left);
      forwardPropagateTree(right);

      String leftCategory = tree.children().get(0).label();
      String rightCategory = tree.children().get(1).label();
      FloatMatrix W = getBinaryTransform(leftCategory, rightCategory);
      classification = getBinaryClassification(leftCategory, rightCategory);

      FloatMatrix leftVector = tree.children().get(0).vector();
      FloatMatrix rightVector = tree.children().get(1).vector();

      FloatMatrix childrenVector = appendBias(leftVector, rightVector);

      if (useFloatTensors) {
        FloatTensor floatT = getBinaryFloatTensor(leftCategory, rightCategory);
        FloatMatrix floatTensorIn = FloatMatrix.concatHorizontally(leftVector, rightVector);
        FloatMatrix floatTensorOut = floatT.bilinearProducts(floatTensorIn);
        nodeVector = activationFunction.apply(W.mmul(childrenVector).add(floatTensorOut));
      } else nodeVector = activationFunction.apply(W.mmul(childrenVector));

    } else {
      throw new AssertionError("Tree not correctly binarized");
    }

    FloatMatrix inputWithBias = appendBias(nodeVector);
    FloatMatrix preAct = classification.mmul(inputWithBias);
    FloatMatrix predictions = outputActivation.apply(preAct);

    tree.setPrediction(predictions);
    tree.setVector(nodeVector);
  }
 /**
  * Add -TMP when not present within an NP
  *
  * @param tree The tree to add temporal info to.
  */
 private void addTMP9(final Tree tree) {
   // do the head chain under it
   Tree ht = headFinder.determineHead(tree);
   // special fix for possessives! -- make noun before head
   if (ht.value().equals("POS")) {
     int j = tree.objectIndexOf(ht);
     if (j > 0) {
       ht = tree.getChild(j - 1);
     }
   }
   // Note: this next bit changes the tree label, rather
   // than creating a new tree node.  Beware!
   if (ht.isPreTerminal()
       || ht.value().startsWith("NP")
       || ht.value().startsWith("PP")
       || ht.value().startsWith("ADVP")) {
     if (!TmpPattern.matcher(ht.value()).matches()) {
       LabelFactory lf = ht.labelFactory();
       // System.err.println("TMP: Changing " + ht.value() + " to " +
       //                   ht.value() + "-TMP");
       ht.setLabel(lf.newLabel(ht.value() + "-TMP"));
     }
     if (ht.value().startsWith("NP")
         || ht.value().startsWith("PP")
         || ht.value().startsWith("ADVP")) {
       addTMP9(ht);
     }
   }
   // do the NPs under it (which may or may not be the head chain
   Tree[] kidlets = tree.children();
   for (int k = 0; k < kidlets.length; k++) {
     ht = kidlets[k];
     LabelFactory lf;
     if (tree.isPrePreTerminal() && !TmpPattern.matcher(ht.value()).matches()) {
       // System.err.println("TMP: Changing " + ht.value() + " to " +
       //                   ht.value() + "-TMP");
       lf = ht.labelFactory();
       // Note: this next bit changes the tree label, rather
       // than creating a new tree node.  Beware!
       ht.setLabel(lf.newLabel(ht.value() + "-TMP"));
     } else if (ht.value().startsWith("NP")) {
       // don't add -TMP twice!
       if (!TmpPattern.matcher(ht.value()).matches()) {
         lf = ht.labelFactory();
         // System.err.println("TMP: Changing " + ht.value() + " to " +
         //                   ht.value() + "-TMP");
         // Note: this next bit changes the tree label, rather
         // than creating a new tree node.  Beware!
         ht.setLabel(lf.newLabel(ht.value() + "-TMP"));
       }
       addTMP9(ht);
     }
   }
 }
Ejemplo n.º 7
0
 List<Tree> prune(List<Tree> treeList, Label label, int start, int end) {
   // get reference tree
   if (treeList.size() == 1) {
     return treeList;
   }
   Tree testTree = treeList.get(0).treeFactory().newTreeNode(label, treeList);
   int goal = Numberer.getGlobalNumberer("states").number(label.value());
   Tree tempTree = parser.extractBestParse(goal, start, end);
   // parser.restoreUnaries(tempTree);
   Tree pcfgTree = debinarizer.transformTree(tempTree);
   Set<Constituent> pcfgConstituents =
       pcfgTree.constituents(new LabeledScoredConstituentFactory());
   // delete child labels that are not in reference but do not cross reference
   List<Tree> prunedChildren = new ArrayList<Tree>();
   int childStart = 0;
   for (int c = 0, numCh = testTree.numChildren(); c < numCh; c++) {
     Tree child = testTree.getChild(c);
     boolean isExtra = true;
     int childEnd = childStart + child.yield().size();
     Constituent childConstituent =
         new LabeledScoredConstituent(childStart, childEnd, child.label(), 0);
     if (pcfgConstituents.contains(childConstituent)) {
       isExtra = false;
     }
     if (childConstituent.crosses(pcfgConstituents)) {
       isExtra = false;
     }
     if (child.isLeaf() || child.isPreTerminal()) {
       isExtra = false;
     }
     if (pcfgTree.yield().size() != testTree.yield().size()) {
       isExtra = false;
     }
     if (!label.value().startsWith("NP^NP")) {
       isExtra = false;
     }
     if (isExtra) {
       System.err.println(
           "Pruning: "
               + child.label()
               + " from "
               + (childStart + start)
               + " to "
               + (childEnd + start));
       System.err.println("Was: " + testTree + " vs " + pcfgTree);
       prunedChildren.addAll(child.getChildrenAsList());
     } else {
       prunedChildren.add(child);
     }
     childStart = childEnd;
   }
   return prunedChildren;
 }
Ejemplo n.º 8
0
 private static void taggedLeafLabels(Tree t, List<CoreLabel> l) {
   if (t.isPreTerminal()) {
     CoreLabel fl = (CoreLabel) t.getChild(0).label();
     fl.set(TagLabelAnnotation.class, t.label());
     l.add(fl);
   } else {
     Tree[] kids = t.children();
     for (int j = 0, n = kids.length; j < n; j++) {
       taggedLeafLabels(kids[j], l);
     }
   }
 }
Ejemplo n.º 9
0
 /** Display a node, implementing Penn Treebank style layout */
 private static <L> void renderTree(
     Tree<L> tree,
     int indent,
     boolean parentLabelNull,
     boolean firstSibling,
     boolean leftSiblingPreTerminal,
     boolean topLevel,
     StringBuilder sb) {
   // the condition for staying on the same line in Penn Treebank
   boolean suppressIndent =
       (parentLabelNull
           || (firstSibling && tree.isPreTerminal())
           || (leftSiblingPreTerminal
               && tree.isPreTerminal()
               && (tree.getLabel() == null || !tree.getLabel().toString().startsWith("CC"))));
   if (suppressIndent) {
     sb.append(' ');
   } else {
     if (!topLevel) {
       sb.append('\n');
     }
     for (int i = 0; i < indent; i++) {
       sb.append("  ");
     }
   }
   if (tree.isLeaf() || tree.isPreTerminal()) {
     renderFlat(tree, sb);
     return;
   }
   sb.append('(');
   sb.append(tree.getLabel());
   renderChildren(
       tree.getChildren(),
       indent + 1,
       tree.getLabel() == null || tree.getLabel().toString() == null,
       sb);
   sb.append(')');
 }
Ejemplo n.º 10
0
 private static <L> void renderChildren(
     List<Tree<L>> children, int indent, boolean parentLabelNull, StringBuilder sb) {
   boolean firstSibling = true;
   boolean leftSibIsPreTerm = true; // counts as true at beginning
   for (Tree<L> child : children) {
     renderTree(child, indent, parentLabelNull, firstSibling, leftSibIsPreTerm, false, sb);
     leftSibIsPreTerm = child.isPreTerminal();
     // CC is a special case
     if (child.getLabel() != null && child.getLabel().toString().startsWith("CC")) {
       leftSibIsPreTerm = false;
     }
     firstSibling = false;
   }
 }
Ejemplo n.º 11
0
 static Tree getPreTerminal(Tree tree, MutableInteger i, int n) {
   if (i.intValue() == n) {
     if (tree.isPreTerminal()) {
       return tree;
     } else {
       return getPreTerminal(tree.children()[0], i, n);
     }
   } else {
     if (tree.isPreTerminal()) {
       i.set(i.intValue() + tree.yield().size());
       return null;
     } else {
       Tree[] kids = tree.children();
       for (int j = 0; j < kids.length; j++) {
         Tree result = getPreTerminal(kids[j], i, n);
         if (result != null) {
           return result;
         }
       }
       return null;
     }
   }
 }
Ejemplo n.º 12
0
  private void backpropDerivativesAndError(
      Tree tree,
      MultiDimensionalMap<String, String, FloatMatrix> binaryTD,
      MultiDimensionalMap<String, String, FloatMatrix> binaryCD,
      MultiDimensionalMap<String, String, FloatTensor> binaryFloatTensorTD,
      Map<String, FloatMatrix> unaryCD,
      Map<String, FloatMatrix> wordVectorD,
      FloatMatrix deltaUp) {
    if (tree.isLeaf()) {
      return;
    }

    FloatMatrix currentVector = tree.vector();
    String category = tree.label();
    category = basicCategory(category);

    // Build a vector that looks like 0,0,1,0,0 with an indicator for the correct class
    FloatMatrix goldLabel = new FloatMatrix(numOuts, 1);
    int goldClass = tree.goldLabel();
    if (goldClass >= 0) {
      goldLabel.put(goldClass, 1.0f);
    }

    Float nodeWeight = classWeights.get(goldClass);
    if (nodeWeight == null) nodeWeight = 1.0f;
    FloatMatrix predictions = tree.prediction();

    // If this is an unlabeled class, set deltaClass to 0.  We could
    // make this more efficient by eliminating various of the below
    // calculations, but this would be the easiest way to handle the
    // unlabeled class
    FloatMatrix deltaClass =
        goldClass >= 0
            ? SimpleBlas.scal(nodeWeight, predictions.sub(goldLabel))
            : new FloatMatrix(predictions.rows, predictions.columns);
    FloatMatrix localCD = deltaClass.mmul(appendBias(currentVector).transpose());

    float error = -(MatrixFunctions.log(predictions).muli(goldLabel).sum());
    error = error * nodeWeight;
    tree.setError(error);

    if (tree.isPreTerminal()) { // below us is a word vector
      unaryCD.put(category, unaryCD.get(category).add(localCD));

      String word = tree.children().get(0).label();
      word = getVocabWord(word);

      FloatMatrix currentVectorDerivative = activationFunction.apply(currentVector);
      FloatMatrix deltaFromClass = getUnaryClassification(category).transpose().mmul(deltaClass);
      deltaFromClass =
          deltaFromClass.get(interval(0, numHidden), interval(0, 1)).mul(currentVectorDerivative);
      FloatMatrix deltaFull = deltaFromClass.add(deltaUp);
      wordVectorD.put(word, wordVectorD.get(word).add(deltaFull));

    } else {
      // Otherwise, this must be a binary node
      String leftCategory = basicCategory(tree.children().get(0).label());
      String rightCategory = basicCategory(tree.children().get(1).label());
      if (combineClassification) {
        unaryCD.put("", unaryCD.get("").add(localCD));
      } else {
        binaryCD.put(
            leftCategory, rightCategory, binaryCD.get(leftCategory, rightCategory).add(localCD));
      }

      FloatMatrix currentVectorDerivative = activationFunction.applyDerivative(currentVector);
      FloatMatrix deltaFromClass =
          getBinaryClassification(leftCategory, rightCategory).transpose().mmul(deltaClass);

      FloatMatrix mult = deltaFromClass.get(interval(0, numHidden), interval(0, 1));
      deltaFromClass = mult.muli(currentVectorDerivative);
      FloatMatrix deltaFull = deltaFromClass.add(deltaUp);

      FloatMatrix leftVector = tree.children().get(0).vector();
      FloatMatrix rightVector = tree.children().get(1).vector();

      FloatMatrix childrenVector = appendBias(leftVector, rightVector);

      // deltaFull 50 x 1, childrenVector: 50 x 2
      FloatMatrix add = binaryTD.get(leftCategory, rightCategory);

      FloatMatrix W_df = deltaFromClass.mmul(childrenVector.transpose());
      binaryTD.put(leftCategory, rightCategory, add.add(W_df));

      FloatMatrix deltaDown;
      if (useFloatTensors) {
        FloatTensor Wt_df = getFloatTensorGradient(deltaFull, leftVector, rightVector);
        binaryFloatTensorTD.put(
            leftCategory,
            rightCategory,
            binaryFloatTensorTD.get(leftCategory, rightCategory).add(Wt_df));
        deltaDown =
            computeFloatTensorDeltaDown(
                deltaFull,
                leftVector,
                rightVector,
                getBinaryTransform(leftCategory, rightCategory),
                getBinaryFloatTensor(leftCategory, rightCategory));
      } else {
        deltaDown = getBinaryTransform(leftCategory, rightCategory).transpose().mmul(deltaFull);
      }

      FloatMatrix leftDerivative = activationFunction.apply(leftVector);
      FloatMatrix rightDerivative = activationFunction.apply(rightVector);
      FloatMatrix leftDeltaDown = deltaDown.get(interval(0, deltaFull.rows), interval(0, 1));
      FloatMatrix rightDeltaDown =
          deltaDown.get(interval(deltaFull.rows, deltaFull.rows * 2), interval(0, 1));
      backpropDerivativesAndError(
          tree.children().get(0),
          binaryTD,
          binaryCD,
          binaryFloatTensorTD,
          unaryCD,
          wordVectorD,
          leftDerivative.mul(leftDeltaDown));
      backpropDerivativesAndError(
          tree.children().get(1),
          binaryTD,
          binaryCD,
          binaryFloatTensorTD,
          unaryCD,
          wordVectorD,
          rightDerivative.mul(rightDeltaDown));
    }
  }
  /**
   * transformTree does all language-specific tree transformations. Any parameterizations should be
   * inside the specific TreebankLangParserParams class.
   */
  @Override
  public Tree transformTree(Tree t, Tree root) {
    if (t == null || t.isLeaf()) {
      return t;
    }

    String parentStr;
    String grandParentStr;
    Tree parent;
    Tree grandParent;
    if (root == null || t.equals(root)) {
      parent = null;
      parentStr = "";
    } else {
      parent = t.parent(root);
      parentStr = parent.label().value();
    }
    if (parent == null || parent.equals(root)) {
      grandParent = null;
      grandParentStr = "";
    } else {
      grandParent = parent.parent(root);
      grandParentStr = grandParent.label().value();
    }

    String baseParentStr = ctlp.basicCategory(parentStr);
    String baseGrandParentStr = ctlp.basicCategory(grandParentStr);

    CoreLabel lab = (CoreLabel) t.label();
    String word = lab.word();
    String tag = lab.tag();
    String baseTag = ctlp.basicCategory(tag);
    String category = lab.value();
    String baseCategory = ctlp.basicCategory(category);

    if (t.isPreTerminal()) { // it's a POS tag
      List<String> leftAunts =
          listBasicCategories(SisterAnnotationStats.leftSisterLabels(parent, grandParent));
      List<String> rightAunts =
          listBasicCategories(SisterAnnotationStats.rightSisterLabels(parent, grandParent));

      // Chinese-specific punctuation splits
      if (chineseSplitPunct && baseTag.equals("PU")) {
        if (ChineseTreebankLanguagePack.chineseDouHaoAcceptFilter().accept(word)) {
          tag = tag + "-DOU";
          // System.out.println("Punct: Split dou hao"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseCommaAcceptFilter().accept(word)) {
          tag = tag + "-COMMA";
          // System.out.println("Punct: Split comma"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseColonAcceptFilter().accept(word)) {
          tag = tag + "-COLON";
          // System.out.println("Punct: Split colon"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseQuoteMarkAcceptFilter().accept(word)) {
          if (chineseSplitPunctLR) {
            if (ChineseTreebankLanguagePack.chineseLeftQuoteMarkAcceptFilter().accept(word)) {
              tag += "-LQUOTE";
            } else {
              tag += "-RQUOTE";
            }
          } else {
            tag = tag + "-QUOTE";
          }
          // System.out.println("Punct: Split quote"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseEndSentenceAcceptFilter().accept(word)) {
          tag = tag + "-ENDSENT";
          // System.out.println("Punct: Split end sent"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseParenthesisAcceptFilter().accept(word)) {
          if (chineseSplitPunctLR) {
            if (ChineseTreebankLanguagePack.chineseLeftParenthesisAcceptFilter().accept(word)) {
              tag += "-LPAREN";
            } else {
              tag += "-RPAREN";
            }
          } else {
            tag += "-PAREN";
            // printlnErr("Just used -PAREN annotation");
            // printlnErr(word);
            // throw new RuntimeException();
          }
          // System.out.println("Punct: Split paren"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseDashAcceptFilter().accept(word)) {
          tag = tag + "-DASH";
          // System.out.println("Punct: Split dash"); // debugging
        } else if (ChineseTreebankLanguagePack.chineseOtherAcceptFilter().accept(word)) {
          tag = tag + "-OTHER";
        } else {
          printlnErr("Unknown punct (you should add it to CTLP): " + tag + " |" + word + "|");
        }
      } else if (chineseSplitDouHao) { // only split DouHao
        if (ChineseTreebankLanguagePack.chineseDouHaoAcceptFilter().accept(word)
            && baseTag.equals("PU")) {
          tag = tag + "-DOU";
        }
      }

      // Chinese-specific POS tag splits (non-punctuation)

      if (tagWordSize) {
        int l = word.length();
        tag += "-" + l + "CHARS";
      }

      if (mergeNNVV && baseTag.equals("NN")) {
        tag = "VV";
      }

      if ((chineseSelectiveTagPA || chineseVerySelectiveTagPA)
          && (baseTag.equals("CC") || baseTag.equals("P"))) {
        tag += "-" + baseParentStr;
      }
      if (chineseSelectiveTagPA && (baseTag.equals("VV"))) {
        tag += "-" + baseParentStr;
      }

      if (markMultiNtag && tag.startsWith("N")) {
        for (int i = 0; i < parent.numChildren(); i++) {
          if (parent.children()[i].label().value().startsWith("N") && parent.children()[i] != t) {
            tag += "=N";
            // System.out.println("Found multi=N rewrite");
          }
        }
      }

      if (markVVsisterIP && baseTag.equals("VV")) {
        boolean seenIP = false;
        for (int i = 0; i < parent.numChildren(); i++) {
          if (parent.children()[i].label().value().startsWith("IP")) {
            seenIP = true;
          }
        }
        if (seenIP) {
          tag += "-IP";
          // System.out.println("Found VV with IP sister"); // testing
        }
      }

      if (markPsisterIP && baseTag.equals("P")) {
        boolean seenIP = false;
        for (int i = 0; i < parent.numChildren(); i++) {
          if (parent.children()[i].label().value().startsWith("IP")) {
            seenIP = true;
          }
        }
        if (seenIP) {
          tag += "-IP";
        }
      }

      if (markADgrandchildOfIP && baseTag.equals("AD") && baseGrandParentStr.equals("IP")) {
        tag += "~IP";
        // System.out.println("Found AD with IP grandparent"); // testing
      }

      if (gpaAD && baseTag.equals("AD")) {
        tag += "~" + baseGrandParentStr;
        // System.out.println("Found AD with grandparent " + grandParentStr); // testing
      }

      if (markPostverbalP && leftAunts.contains("VV") && baseTag.equals("P")) {
        // System.out.println("Found post-verbal P");
        tag += "^=lVV";
      }

      // end Chinese-specific tag splits

      Label label = new CategoryWordTag(tag, word, tag);
      t.setLabel(label);
    } else {
      // it's a phrasal category
      Tree[] kids = t.children();

      // Chinese-specific category splits
      List<String> leftSis = listBasicCategories(SisterAnnotationStats.leftSisterLabels(t, parent));
      List<String> rightSis =
          listBasicCategories(SisterAnnotationStats.rightSisterLabels(t, parent));

      if (paRootDtr && baseParentStr.equals("ROOT")) {
        category += "^ROOT";
      }

      if (markIPsisterBA && baseCategory.equals("IP")) {
        if (leftSis.contains("BA")) {
          category += "=BA";
          // System.out.println("Found IP sister of BA");
        }
      }

      if (dominatesV && hasV(t.preTerminalYield())) {
        // mark categories containing a verb
        category += "-v";
      }

      if (markIPsisterVVorP && baseCategory.equals("IP")) {
        // todo: cdm: is just looking for "P" here selective enough??
        if (leftSis.contains("VV") || leftSis.contains("P")) {
          category += "=VVP";
        }
      }

      if (markIPsisDEC && baseCategory.equals("IP")) {
        if (rightSis.contains("DEC")) {
          category += "=DEC";
          // System.out.println("Found prenominal IP");
        }
      }

      if (baseCategory.equals("VP")) {
        // cdm 2008: this used to just check that it startsWith("VP"), but
        // I think that was bad because it also matched VPT verb compounds
        if (chineseSplitVP == 3) {
          boolean hasCC = false;
          boolean hasPU = false;
          boolean hasLexV = false;
          for (Tree kid : kids) {
            if (kid.label().value().startsWith("CC")) {
              hasCC = true;
            } else if (kid.label().value().startsWith("PU")) {
              hasPU = true;
            } else if (StringUtils.lookingAt(
                kid.label().value(), "(V[ACEV]|VCD|VCP|VNV|VPT|VRD|VSB)")) {
              hasLexV = true;
            }
          }
          if (hasCC || (hasPU && !hasLexV)) {
            category += "-CRD";
            // System.out.println("Found coordinate VP"); // testing
          } else if (hasLexV) {
            category += "-COMP";
            // System.out.println("Found complementing VP"); // testing
          } else {
            category += "-ADJT";
            // System.out.println("Found adjoining VP"); // testing
          }
        } else if (chineseSplitVP >= 1) {
          boolean hasBA = false;
          for (Tree kid : kids) {
            if (kid.label().value().startsWith("BA")) {
              hasBA = true;
            } else if (chineseSplitVP == 2 && tlp.basicCategory(kid.label().value()).equals("VP")) {
              for (Tree kidkid : kid.children()) {
                if (kidkid.label().value().startsWith("BA")) {
                  hasBA = true;
                }
              }
            }
          }
          if (hasBA) {
            category += "-BA";
          }
        }
      }

      if (markVPadjunct && baseParentStr.equals("VP")) {
        // cdm 2008: This used to use startsWith("VP") but changed to baseCat
        Tree[] sisters = parent.children();
        boolean hasVPsister = false;
        boolean hasCC = false;
        boolean hasPU = false;
        boolean hasLexV = false;
        for (Tree sister : sisters) {
          if (tlp.basicCategory(sister.label().value()).equals("VP")) {
            hasVPsister = true;
          }
          if (sister.label().value().startsWith("CC")) {
            hasCC = true;
          }
          if (sister.label().value().startsWith("PU")) {
            hasPU = true;
          }
          if (StringUtils.lookingAt(sister.label().value(), "(V[ACEV]|VCD|VCP|VNV|VPT|VRD|VSB)")) {
            hasLexV = true;
          }
        }
        if (hasVPsister && !(hasCC || hasPU || hasLexV)) {
          category += "-VPADJ";
          // System.out.println("Found adjunct of VP"); // testing
        }
      }

      if (markNPmodNP && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
        if (rightSis.contains("NP")) {
          category += "=MODIFIERNP";
          // System.out.println("Found NP modifier of NP"); // testing
        }
      }

      if (markModifiedNP && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
        if (rightSis.isEmpty()
            && (leftSis.contains("ADJP")
                || leftSis.contains("NP")
                || leftSis.contains("DNP")
                || leftSis.contains("QP")
                || leftSis.contains("CP")
                || leftSis.contains("PP"))) {
          category += "=MODIFIEDNP";
          // System.out.println("Found modified NP"); // testing
        }
      }

      if (markNPconj && baseCategory.equals("NP") && baseParentStr.equals("NP")) {
        if (rightSis.contains("CC")
            || rightSis.contains("PU")
            || leftSis.contains("CC")
            || leftSis.contains("PU")) {
          category += "=CONJ";
          // System.out.println("Found NP conjunct"); // testing
        }
      }

      if (markIPconj && baseCategory.equals("IP") && baseParentStr.equals("IP")) {
        Tree[] sisters = parent.children();
        boolean hasCommaSis = false;
        boolean hasIPSis = false;
        for (Tree sister : sisters) {
          if (ctlp.basicCategory(sister.label().value()).equals("PU")
              && ChineseTreebankLanguagePack.chineseCommaAcceptFilter()
                  .accept(sister.children()[0].label().toString())) {
            hasCommaSis = true;
            // System.out.println("Found CommaSis"); // testing
          }
          if (ctlp.basicCategory(sister.label().value()).equals("IP") && sister != t) {
            hasIPSis = true;
          }
        }
        if (hasCommaSis && hasIPSis) {
          category += "-CONJ";
          // System.out.println("Found IP conjunct"); // testing
        }
      }

      if (unaryIP && baseCategory.equals("IP") && t.numChildren() == 1) {
        category += "-U";
        // System.out.println("Found unary IP"); //testing
      }
      if (unaryCP && baseCategory.equals("CP") && t.numChildren() == 1) {
        category += "-U";
        // System.out.println("Found unary CP"); //testing
      }

      if (splitBaseNP && baseCategory.equals("NP")) {
        if (t.isPrePreTerminal()) {
          category = category + "-B";
        }
      }

      // if (Test.verbose) printlnErr(baseCategory + " " + leftSis.toString()); //debugging

      if (markPostverbalPP && leftSis.contains("VV") && baseCategory.equals("PP")) {
        // System.out.println("Found post-verbal PP");
        category += "=lVV";
      }

      if ((markADgrandchildOfIP || gpaAD)
          && listBasicCategories(SisterAnnotationStats.kidLabels(t)).contains("AD")) {
        category += "^ADVP";
      }

      if (markCC) {
        // was: for (int i = 0; i < kids.length; i++) {
        // This second version takes an idea from Collins: don't count
        // marginal conjunctions which don't conjoin 2 things.
        for (int i = 1; i < kids.length - 1; i++) {
          String cat2 = kids[i].label().value();
          if (cat2.startsWith("CC")) {
            category += "-CC";
          }
        }
      }

      Label label = new CategoryWordTag(category, word, tag);
      t.setLabel(label);
    }
    return t;
  }