/**
* Returns the index of <code>daughter</code> in <code>parent</code> by ==. Returns -1 if <code>
* daughter</code> not found.
*/
public static int objectEqualityIndexOf(Tree parent, Tree daughter) {
for (int i = 0; i < parent.children().length; i++) {
if (daughter == parent.children()[i]) {
return i;
}
}
return -1;
}
/**
* returns the syntactic category of the tree as a list of the syntactic categories of the mother
* and the daughters
*/
public static List<String> localTreeAsCatList(Tree t) {
List<String> l = new ArrayList<String>(t.children().length + 1);
l.add(t.label().value());
for (int i = 0; i < t.children().length; i++) {
l.add(t.children()[i].label().value());
}
return l;
}
public FloatMatrix getWForNode(Tree node) {
if (node.children().size() == 2) {
String leftLabel = node.children().get(0).value();
String leftBasic = basicCategory(leftLabel);
String rightLabel = node.children().get(1).value();
String rightBasic = basicCategory(rightLabel);
return binaryTransform.get(leftBasic, rightBasic);
} else if (node.children().size() == 1) {
throw new AssertionError("No unary transform matrices, only unary classification");
} else {
throw new AssertionError("Unexpected tree children size of " + node.children().size());
}
}
protected Rule ltToRule(Tree lt) {
if (lt.children().length == 1) {
UnaryRule ur = new UnaryRule();
ur.parent = stateNumberer.number(lt.label().value());
ur.child = stateNumberer.number(lt.children()[0].label().value());
return ur;
} else {
BinaryRule br = new BinaryRule();
br.parent = stateNumberer.number(lt.label().value());
br.leftChild = stateNumberer.number(lt.children()[0].label().value());
br.rightChild = stateNumberer.number(lt.children()[1].label().value());
return br;
}
}
private static void leafLabels(Tree t, List<Label> l) {
if (t.isLeaf()) {
l.add(t.label());
} else {
Tree[] kids = t.children();
for (int j = 0, n = kids.length; j < n; j++) {
leafLabels(kids[j], l);
}
}
}
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);
}
}
}
public static <A> void printTree(Tree<A> tree, PrintStream ps)
{
ps.printf("(%s (", tree.value());
for (Tree<A> t : tree.children())
{
printTree(t, ps);
ps.print(",");
}
ps.print("))");
}
/**
* 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);
}
}
}
private static boolean includesEmptyNPSubj(Tree t) {
if (t == null) {
return false;
}
Tree[] kids = t.children();
if (kids == null) {
return false;
}
boolean foundNullSubj = false;
for (Tree kid : kids) {
Tree[] kidkids = kid.children();
if (NPSbjPattern.matcher(kid.value()).matches()) {
kid.setValue("NP");
if (kidkids != null && kidkids.length == 1 && kidkids[0].value().equals("-NONE-")) {
// only set flag, since there are 2 a couple of times (errors)
foundNullSubj = true;
}
}
}
return foundNullSubj;
}
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);
}
}
}
/**
* 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);
}
private static int treeToLatexHelper(
Tree t, StringBuilder c, StringBuilder h, int n, int nextN, int indent) {
StringBuilder sb = new StringBuilder();
for (int i = 0; i < indent; i++) sb.append(" ");
h.append('\n').append(sb);
h.append("{\\")
.append(t.isLeaf() ? "" : "n")
.append("tnode{z")
.append(n)
.append("}{")
.append(t.label())
.append('}');
if (!t.isLeaf()) {
for (int k = 0; k < t.children().length; k++) {
h.append(", ");
c.append("\\nodeconnect{z").append(n).append("}{z").append(nextN).append("}\n");
nextN = treeToLatexHelper(t.children()[k], c, h, nextN, nextN + 1, indent + 1);
}
}
h.append('}');
return nextN;
}
// TODO all this is very memory inefficient and will lead to stack overflows for very deep trees
public static ObjectNode toJsonTree(ObjectMapper objectMapper, Tree<? extends JsonNode> tree)
{
final ObjectNode node = objectMapper.createObjectNode();
node.put("_value", tree.value());
final ArrayNode children = objectMapper.createArrayNode();
for (Tree<? extends JsonNode> child : tree.children())
children.add(toJsonTree(objectMapper, child));
node.put("_children", children);
return node;
}
protected void tallyTree(Tree t, LinkedList<String> parents) {
// traverse tree, building parent list
String str = t.label().value();
boolean strIsPassive = (str.indexOf('@') == -1);
if (strIsPassive) {
parents.addFirst(str);
}
if (!t.isLeaf()) {
if (!t.children()[0].isLeaf()) {
tallyInternalNode(t, parents);
for (int c = 0; c < t.children().length; c++) {
Tree child = t.children()[c];
tallyTree(child, parents);
}
} else {
tagNumberer.number(t.label().value());
}
}
if (strIsPassive) {
parents.removeFirst();
}
}
/**
* traceTo() values that are contained in the tree are assigned to new objects. traceTo() values
* that are not contained in the tree are given the old value.
*/
public static void fixEmptyTreeLeafs(Tree t, Map<Tree, Tree> newToOld, Map<Tree, Tree> oldToNew) {
Tree[] kids = t.children();
for (int i = 0, n = kids.length; i < n; i++) {
fixEmptyTreeLeafs(kids[i], newToOld, oldToNew);
}
if (t instanceof EmptyTreeLeaf) {
EmptyTreeLeaf oldT = (EmptyTreeLeaf) newToOld.get(t);
((EmptyTreeLeaf) t).setEmptyType(oldT.emptyType());
Tree oldTraceTo = oldT.traceTo();
Tree newTraceTo = oldToNew.get(oldTraceTo);
if (newTraceTo != null) ((EmptyTreeLeaf) t).setTraceTo(newTraceTo);
else ((EmptyTreeLeaf) t).setTraceTo(oldTraceTo);
}
}
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;
}
}
}
/** replaces all instances (by ==) of node with node1. Doesn't affect the node t itself */
public static void replaceNode(Tree node, Tree node1, Tree t) {
if (t.isLeaf()) return;
Tree[] kids = t.children();
List<Tree> newKids = new ArrayList<Tree>(kids.length);
for (int i = 0, n = kids.length; i < n; i++) {
if (kids[i] != node) {
newKids.add(kids[i]);
replaceNode(node, node1, kids[i]);
} else {
newKids.add(node1);
}
}
t.setChildren(newKids);
}
private static int treeToLatexEvenHelper(
Tree t,
StringBuilder c,
StringBuilder h,
int n,
int nextN,
int indent,
int curDepth,
int maxDepth) {
StringBuilder sb = new StringBuilder();
for (int i = 0; i < indent; i++) sb.append(" ");
h.append('\n').append(sb);
int tDepth = t.depth();
if (tDepth == 0 && tDepth + curDepth < maxDepth) {
for (int pad = 0; pad < maxDepth - tDepth - curDepth; pad++) {
h.append("{\\ntnode{pad}{}, ");
}
}
h.append("{\\ntnode{z").append(n).append("}{").append(t.label()).append('}');
if (!t.isLeaf()) {
for (int k = 0; k < t.children().length; k++) {
h.append(", ");
c.append("\\nodeconnect{z").append(n).append("}{z").append(nextN).append("}\n");
nextN =
treeToLatexEvenHelper(
t.children()[k], c, h, nextN, nextN + 1, indent + 1, curDepth + 1, maxDepth);
}
}
if (tDepth == 0 && tDepth + curDepth < maxDepth) {
for (int pad = 0; pad < maxDepth - tDepth - curDepth; pad++) {
h.append('}');
}
}
h.append('}');
return nextN;
}
public static ArrayList<ArrayList<TaggedWord>> getPhrases(Tree parse, int phraseSizeLimit) {
ArrayList<ArrayList<TaggedWord>> newList = new ArrayList<ArrayList<TaggedWord>>();
List<Tree> leaves = parse.getLeaves();
if (leaves.size() <= phraseSizeLimit) {
// ArrayList<TaggedWord> phraseElements = PreprocessPhrase(parse.taggedYield());
ArrayList<TaggedWord> phraseElements = Preprocess(parse.taggedYield());
if (phraseElements.size() > 0) newList.add(phraseElements);
} else {
Tree[] childrenNodes = parse.children();
for (int i = 0; i < childrenNodes.length; i++) {
Tree currentParse = childrenNodes[i];
newList.addAll(getPhrases(currentParse, phraseSizeLimit));
}
}
return newList;
}
static boolean rightEdge(Tree t, Tree t1, MutableInteger i) {
if (t == t1) {
return true;
} else if (t1.isLeaf()) {
int j = t1.yield().size(); // so that empties don't add size
i.set(i.intValue() - j);
return false;
} else {
Tree[] kids = t1.children();
for (int j = kids.length - 1; j >= 0; j--) {
if (rightEdge(t, kids[j], i)) {
return true;
}
}
return false;
}
}
public static Tree copyHelper(Tree t, Map<Tree, Tree> newToOld, Map<Tree, Tree> oldToNew) {
Tree[] kids = t.children();
Tree[] newKids = new Tree[kids.length];
for (int i = 0, n = kids.length; i < n; i++) {
newKids[i] = copyHelper(kids[i], newToOld, oldToNew);
}
TreeFactory tf = t.treeFactory();
if (kids.length == 0) {
Tree newLeaf = tf.newLeaf(t.label());
newToOld.put(newLeaf, t);
oldToNew.put(newLeaf, t);
return newLeaf;
}
Tree newNode = tf.newTreeNode(t.label(), Arrays.asList(newKids));
newToOld.put(newNode, t);
oldToNew.put(t, newNode);
return newNode;
}
/**
* Map a function over a tree
*
* @param fn Function
* @param tree Tree of {@code A}'s
* @return Tree of {@code B}'s
*/
public static <A,B> Tree<B> map(final Function<A,B> fn, Tree<A> tree)
{
final B value = fn.apply(tree.value());
if (isLeaf(tree))
return new ImmutableTree<B>(value);
else
{
final Function<Tree<A>,Tree<B>> tmap = new Function<Tree<A>,Tree<B>>()
{
public Tree<B> apply(Tree<A> tree)
{
return map(fn, tree);
}
};
final Iterable<Tree<B>> tb = Iterables.transform(tree.children(), tmap);
return new ImmutableTree<B>(value, tb);
}
}
public FloatMatrix getClassWForNode(Tree node) {
if (combineClassification) {
return unaryClassification.get("");
} else if (node.children().size() == 2) {
String leftLabel = node.children().get(0).value();
String leftBasic = basicCategory(leftLabel);
String rightLabel = node.children().get(1).value();
String rightBasic = basicCategory(rightLabel);
return binaryClassification.get(leftBasic, rightBasic);
} else if (node.children().size() == 1) {
String unaryLabel = node.children().get(0).value();
String unaryBasic = basicCategory(unaryLabel);
return unaryClassification.get(unaryBasic);
} else {
throw new AssertionError("Unexpected tree children size of " + node.children().size());
}
}
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));
}
}
public static final <_> boolean isLeaf(final Tree<_> tree)
{
return Iterables.isEmpty(tree.children());
}
/**
* 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;
}
