/**
* Determine if the given tree contains a leaf which matches the part-of-speech and lexical
* criteria.
*
* @param pos Regular expression to match part of speech (may be null, in which case any POS is
* allowed)
* @param pos Regular expression to match word (may be null, in which case any word is allowed)
*/
public static boolean shouldPrintTree(Tree tree, Pattern pos, Pattern word) {
for (Tree t : tree) {
if (t.isPreTerminal()) {
CoreLabel label = (CoreLabel) t.label();
String tpos = label.value();
Tree wordNode = t.firstChild();
CoreLabel wordLabel = (CoreLabel) wordNode.label();
String tword = wordLabel.value();
if ((pos == null || pos.matcher(tpos).find())
&& (word == null || word.matcher(tword).find())) return true;
}
}
return false;
}
private static String toString(final List<CoreLabel> lineage) {
StringBuilder sb = new StringBuilder();
for (CoreLabel cl : lineage) {
sb.append(cl.value());
sb.append(" <-- ");
}
return sb.toString();
}
/**
* Depth-first (post-order) search through the tree, recording the stack state as the lineage
* every time a terminal is reached.
*
* <p>This implementation uses the Index annotation to store depth. If CoreLabels are not present
* in the trees (or at least something that implements HasIndex), an exception will result.
*
* @param t The tree
* @return A list of lineages
*/
private static List<List<CoreLabel>> makeLineages(final Tree t) {
if (t == null) return null;
((HasIndex) t.label()).setIndex(0);
final Stack<Tree> treeStack = new Stack<>();
treeStack.push(t);
final Stack<CoreLabel> labelStack = new Stack<>();
CoreLabel rootLabel = new CoreLabel(t.label());
rootLabel.setIndex(0);
labelStack.push(rootLabel);
final List<List<CoreLabel>> lineages = new ArrayList<>();
while (!treeStack.isEmpty()) {
Tree node = treeStack.pop();
int nodeDepth = ((HasIndex) node.label()).index();
while (!labelStack.isEmpty() && labelStack.peek().index() != nodeDepth - 1) labelStack.pop();
if (node.isPreTerminal()) {
List<CoreLabel> lin = new ArrayList<>(labelStack);
lineages.add(lin);
} else {
for (Tree kid : node.children()) {
((HasIndex) kid.label()).setIndex(nodeDepth + 1);
treeStack.push(kid);
}
CoreLabel nodeLabel = new CoreLabel(node.label());
nodeLabel.setIndex(nodeDepth);
labelStack.add(nodeLabel);
}
}
if (DEBUG) {
System.out.println("Lineages:");
for (List<CoreLabel> lin : lineages) {
for (CoreLabel cl : lin) System.out.print(cl.value() + " <- ");
System.out.println();
}
}
return lineages;
}
public DataBag exec(Tuple input) throws IOException {
if (input == null || input.size() < 1 || input.isNull(0)) return null;
// Output bag
DataBag bagOfTokens = bagFactory.newDefaultBag();
StringReader textInput = new StringReader(input.get(0).toString());
PTBTokenizer ptbt = new PTBTokenizer(textInput, new CoreLabelTokenFactory(), "");
for (CoreLabel label; ptbt.hasNext(); ) {
label = (CoreLabel) ptbt.next();
if (label.value().length() > 2) {
System.err.println(label.toString());
Tuple termText = tupleFactory.newTuple(label.word());
bagOfTokens.add(termText);
}
}
return bagOfTokens;
}
/**
* 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;
}