/**
* 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);
}
}
}
// now overally complex so it deals with coordinations. Maybe change this class to use tregrex?
private boolean hasPassiveProgressiveAuxiliary(Tree[] kids, HashSet<String> verbalSet) {
if (DEBUG) {
System.err.println("Checking for passive/progressive auxiliary");
}
boolean foundPassiveVP = false;
boolean foundPassiveAux = false;
for (Tree kid : kids) {
if (DEBUG) {
System.err.println(" checking in " + kid);
}
if (kid.isPreTerminal()) {
Label kidLabel = kid.label();
String tag = null;
if (kidLabel instanceof HasTag) {
tag = ((HasTag) kidLabel).tag();
}
if (tag == null) {
tag = kid.value();
}
Label wordLabel = kid.firstChild().label();
String word = null;
if (wordLabel instanceof HasWord) {
word = ((HasWord) wordLabel).word();
}
if (word == null) {
word = wordLabel.value();
}
if (DEBUG) {
System.err.println("Checking " + kid.value() + " head is " + word + '/' + tag);
}
String lcWord = word.toLowerCase();
if (verbalTags.contains(tag) && verbalSet.contains(lcWord)) {
if (DEBUG) {
System.err.println("hasPassiveProgressiveAuxiliary found passive aux");
}
foundPassiveAux = true;
}
} else if (kid.isPhrasal()) {
Label kidLabel = kid.label();
String cat = null;
if (kidLabel instanceof HasCategory) {
cat = ((HasCategory) kidLabel).category();
}
if (cat == null) {
cat = kid.value();
}
if (!cat.startsWith("VP")) {
continue;
}
if (DEBUG) {
System.err.println("hasPassiveProgressiveAuxiliary found VP");
}
Tree[] kidkids = kid.children();
boolean foundParticipleInVp = false;
for (Tree kidkid : kidkids) {
if (DEBUG) {
System.err.println(" hasPassiveProgressiveAuxiliary examining " + kidkid);
}
if (kidkid.isPreTerminal()) {
Label kidkidLabel = kidkid.label();
String tag = null;
if (kidkidLabel instanceof HasTag) {
tag = ((HasTag) kidkidLabel).tag();
}
if (tag == null) {
tag = kidkid.value();
}
// we allow in VBD because of frequent tagging mistakes
if ("VBN".equals(tag) || "VBG".equals(tag) || "VBD".equals(tag)) {
foundPassiveVP = true;
if (DEBUG) {
System.err.println("hasPassiveAuxiliary found VBN/VBG/VBD VP");
}
break;
} else if ("CC".equals(tag) && foundParticipleInVp) {
foundPassiveVP = true;
if (DEBUG) {
System.err.println(
"hasPassiveAuxiliary [coordination] found (VP (VP[VBN/VBG/VBD] CC");
}
break;
}
} else if (kidkid.isPhrasal()) {
String catcat = null;
if (kidLabel instanceof HasCategory) {
catcat = ((HasCategory) kidLabel).category();
}
if (catcat == null) {
catcat = kid.value();
}
if ("VP".equals(catcat)) {
if (DEBUG) {
System.err.println("hasPassiveAuxiliary found (VP (VP)), recursing");
}
foundParticipleInVp = vpContainsParticiple(kidkid);
} else if (("CONJP".equals(catcat) || "PRN".equals(catcat))
&& foundParticipleInVp) { // occasionally get PRN in CONJ-like structures
foundPassiveVP = true;
if (DEBUG) {
System.err.println(
"hasPassiveAuxiliary [coordination] found (VP (VP[VBN/VBG/VBD] CONJP");
}
break;
}
}
}
}
if (foundPassiveAux && foundPassiveVP) {
break;
}
} // end for (Tree kid : kids)
if (DEBUG) {
System.err.println(
"hasPassiveProgressiveAuxiliary returns " + (foundPassiveAux && foundPassiveVP));
}
return foundPassiveAux && foundPassiveVP;
}
/**
* Normalize a whole tree -- one can assume that this is the root. This implementation deletes
* empty elements (ones with nonterminal tag label '-NONE-') from the tree.
*/
@Override
public Tree normalizeWholeTree(Tree tree, TreeFactory tf) {
TreeTransformer transformer1 =
new TreeTransformer() {
@Override
public Tree transformTree(Tree t) {
if (doSGappedStuff) {
String lab = t.label().value();
if (lab.equals("S") && includesEmptyNPSubj(t)) {
LabelFactory lf = t.label().labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
t.setLabel(lf.newLabel(t.label().value() + "-G"));
}
}
return t;
}
};
Filter<Tree> subtreeFilter =
new Filter<Tree>() {
private static final long serialVersionUID = -7250433816896327901L;
@Override
public boolean accept(Tree t) {
Tree[] kids = t.children();
Label l = t.label();
// The special Switchboard non-terminals clause.
// Note that it deletes IP which other Treebanks might use!
if ("RS".equals(t.label().value())
|| "RM".equals(t.label().value())
|| "IP".equals(t.label().value())
|| "CODE".equals(t.label().value())) {
return false;
}
if ((l != null)
&& l.value() != null
&& (l.value().equals("-NONE-"))
&& !t.isLeaf()
&& kids.length == 1
&& kids[0].isLeaf()) {
// Delete empty/trace nodes (ones marked '-NONE-')
return false;
}
return true;
}
};
Filter<Tree> nodeFilter =
new Filter<Tree>() {
private static final long serialVersionUID = 9000955019205336311L;
@Override
public boolean accept(Tree t) {
if (t.isLeaf() || t.isPreTerminal()) {
return true;
}
// The special switchboard non-terminals clause. Try keeping EDITED for now....
// if ("EDITED".equals(t.label().value())) {
// return false;
// }
if (t.numChildren() != 1) {
return true;
}
if (t.label() != null
&& t.label().value() != null
&& t.label().value().equals(t.children()[0].label().value())) {
return false;
}
return true;
}
};
TreeTransformer transformer2 =
new TreeTransformer() {
@Override
public Tree transformTree(Tree t) {
if (temporalAnnotation == TEMPORAL_ANY_TMP_PERCOLATED) {
String lab = t.label().value();
if (TmpPattern.matcher(lab).matches()) {
Tree oldT = t;
Tree ht;
do {
ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.label().value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.getChild(j - 1);
}
}
LabelFactory lf = ht.label().labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-TMP"));
oldT = ht;
} while (!ht.isPreTerminal());
if (lab.startsWith("PP")) {
ht = headFinder.determineHead(t);
// look to right
int j = t.objectIndexOf(ht);
int sz = t.children().length;
if (j + 1 < sz) {
ht = t.getChild(j + 1);
}
if (ht.label().value().startsWith("NP")) {
while (!ht.isLeaf()) {
LabelFactory lf = ht.label().labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-TMP"));
ht = headFinder.determineHead(ht);
}
}
}
}
} else if (temporalAnnotation == TEMPORAL_ALL_TERMINALS) {
String lab = t.label().value();
if (NPTmpPattern.matcher(lab).matches()) {
Tree ht;
ht = headFinder.determineHead(t);
if (ht.isPreTerminal()) {
// change all tags to -TMP
LabelFactory lf = ht.label().labelFactory();
Tree[] kids = t.children();
for (Tree kid : kids) {
if (kid.isPreTerminal()) {
// Note: this changes the tree label, rather
// than creating a new tree node. Beware!
kid.setLabel(lf.newLabel(kid.value() + "-TMP"));
}
}
} else {
Tree oldT = t;
do {
ht = headFinder.determineHead(oldT);
oldT = ht;
} while (!ht.isPreTerminal());
LabelFactory lf = ht.label().labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-TMP"));
}
}
} else if (temporalAnnotation == TEMPORAL_ALL_NP) {
String lab = t.label().value();
if (NPTmpPattern.matcher(lab).matches()) {
Tree oldT = t;
Tree ht;
do {
ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.label().value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.getChild(j - 1);
}
}
if (ht.isPreTerminal() || ht.value().startsWith("NP")) {
LabelFactory lf = ht.labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-TMP"));
oldT = ht;
}
} while (ht.value().startsWith("NP"));
}
} else if (temporalAnnotation == TEMPORAL_ALL_NP_AND_PP
|| temporalAnnotation == TEMPORAL_NP_AND_PP_WITH_NP_HEAD
|| temporalAnnotation == TEMPORAL_ALL_NP_EVEN_UNDER_PP) {
// also allow chain to start with PP
String lab = t.value();
if (NPTmpPattern.matcher(lab).matches() || PPTmpPattern.matcher(lab).matches()) {
Tree oldT = t;
do {
Tree ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.getChild(j - 1);
}
} else if ((temporalAnnotation == TEMPORAL_NP_AND_PP_WITH_NP_HEAD
|| temporalAnnotation == TEMPORAL_ALL_NP_EVEN_UNDER_PP)
&& (ht.value().equals("IN") || ht.value().equals("TO"))) {
// change the head to be NP if possible
Tree[] kidlets = oldT.children();
for (int k = kidlets.length - 1; k > 0; k--) {
if (kidlets[k].value().startsWith("NP")) {
ht = kidlets[k];
}
}
}
LabelFactory lf = ht.labelFactory();
// Note: this next bit changes the tree label, rather
// than creating a new tree node. Beware!
if (ht.isPreTerminal() || ht.value().startsWith("NP")) {
ht.setLabel(lf.newLabel(ht.value() + "-TMP"));
}
if (temporalAnnotation == TEMPORAL_ALL_NP_EVEN_UNDER_PP
&& oldT.value().startsWith("PP")) {
oldT.setLabel(lf.newLabel(tlp.basicCategory(oldT.value())));
}
oldT = ht;
} while (oldT.value().startsWith("NP") || oldT.value().startsWith("PP"));
}
} else if (temporalAnnotation == TEMPORAL_ALL_NP_PP_ADVP) {
// also allow chain to start with PP or ADVP
String lab = t.value();
if (NPTmpPattern.matcher(lab).matches()
|| PPTmpPattern.matcher(lab).matches()
|| ADVPTmpPattern.matcher(lab).matches()) {
Tree oldT = t;
do {
Tree ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.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")) {
LabelFactory lf = ht.labelFactory();
ht.setLabel(lf.newLabel(ht.value() + "-TMP"));
}
oldT = ht;
} while (oldT.value().startsWith("NP"));
}
} else if (temporalAnnotation == TEMPORAL_9) {
// also allow chain to start with PP or ADVP
String lab = t.value();
if (NPTmpPattern.matcher(lab).matches()
|| PPTmpPattern.matcher(lab).matches()
|| ADVPTmpPattern.matcher(lab).matches()) {
// System.err.println("TMP: Annotating " + t);
addTMP9(t);
}
} else if (temporalAnnotation == TEMPORAL_ACL03PCFG) {
String lab = t.label().value();
if (lab != null && NPTmpPattern.matcher(lab).matches()) {
Tree oldT = t;
Tree ht;
do {
ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.label().value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.getChild(j - 1);
}
}
oldT = ht;
} while (!ht.isPreTerminal());
if (!onlyTagAnnotateNstar || ht.label().value().startsWith("N")) {
LabelFactory lf = ht.label().labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-TMP"));
}
}
}
if (doAdverbialNP) {
String lab = t.value();
if (NPAdvPattern.matcher(lab).matches()) {
Tree oldT = t;
Tree ht;
do {
ht = headFinder.determineHead(oldT);
// special fix for possessives! -- make noun before head
if (ht.label().value().equals("POS")) {
int j = oldT.objectIndexOf(ht);
if (j > 0) {
ht = oldT.getChild(j - 1);
}
}
if (ht.isPreTerminal() || ht.value().startsWith("NP")) {
LabelFactory lf = ht.labelFactory();
// Note: this changes the tree label, rather than
// creating a new tree node. Beware!
ht.setLabel(lf.newLabel(ht.label().value() + "-ADV"));
oldT = ht;
}
} while (ht.value().startsWith("NP"));
}
}
return t;
}
};
// if there wasn't an empty nonterminal at the top, but an S, wrap it.
if (tree.label().value().equals("S")) {
tree = tf.newTreeNode("ROOT", Collections.singletonList(tree));
}
// repair for the phrasal VB in Switchboard (PTB version 3) that should be a VP
for (Tree subtree : tree) {
if (subtree.isPhrasal() && "VB".equals(subtree.label().value())) {
subtree.setValue("VP");
}
}
tree = tree.transform(transformer1);
if (tree == null) {
return null;
}
tree = tree.prune(subtreeFilter, tf);
if (tree == null) {
return null;
}
tree = tree.spliceOut(nodeFilter, tf);
if (tree == null) {
return null;
}
return tree.transform(transformer2, tf);
}