/**
* Given a tree t, if this tree contains a QP of the form QP (RB IN CD|DT ...) well over, more
* than QP (JJR IN CD|DT ...) fewer than QP (IN JJS CD|DT ...) at least QP (... CC ...) between 5
* and 10 it will transform it
*/
private static void doTransform(Tree t) {
if (t.value().startsWith("QP")) {
// look at the children
List<Tree> children = t.getChildrenAsList();
if (children.size() >= 3 && children.get(0).isPreTerminal()) {
// go through the children and check if they match the structure we want
String child1 = children.get(0).value();
String child2 = children.get(1).value();
String child3 = children.get(2).value();
if ((child3.startsWith("CD") || child3.startsWith("DT"))
&& (child1.startsWith("RB") || child1.startsWith("JJ") || child1.startsWith("IN"))
&& (child2.startsWith("IN") || child2.startsWith("JJ"))) {
transformQP(t);
children = t.getChildrenAsList();
}
}
// If the children include a CC, we split that into left and
// right subtrees with the CC in the middle so the headfinders
// have an easier time interpreting the tree later on
if (children.size() >= 3) {
boolean flat = true;
for (int i = 0; i < children.size(); ++i) {
if (!children.get(i).isPreTerminal()) {
flat = false;
break;
}
}
if (flat) {
for (int i = 1; i < children.size() - 1; ++i) {
if (children.get(i).value().startsWith("CC")) {
transformCC(
t,
children.subList(0, i),
children.get(i),
children.subList(i + 1, children.size()));
break;
}
}
}
}
/* --- to be written or deleted
} else if (t.value().startsWith("NP")) {
//look at the children
List<Tree> children = t.getChildrenAsList();
if (children.size() >= 3) {
}
---- */
} else if (t.isPhrasal()) {
for (Tree child : t.children()) {
doTransform(child);
}
}
}
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);
}
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;
}
private static void transformQP(Tree t) {
List<Tree> children = t.getChildrenAsList();
TreeFactory tf = t.treeFactory();
LabelFactory lf = t.label().labelFactory();
// create the new XS having the first two children of the QP
Tree left = tf.newTreeNode(lf.newLabel("XS"), null);
for (int i = 0; i < 2; i++) {
left.addChild(children.get(i));
}
// remove all the two first children of t before
for (int i = 0; i < 2; i++) {
t.removeChild(0);
}
// add XS as the first child
t.addChild(0, left);
}