Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
package edu.stanford.nlp.trees;
import static java.lang.System.err;
import java.io.PrintWriter;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.List;
import edu.stanford.nlp.ling.Label;
import edu.stanford.nlp.ling.TaggedWord;
import edu.stanford.nlp.ling.WordTag;
import edu.stanford.nlp.process.Morphology;
import edu.stanford.nlp.util.ArrayUtils;
import edu.stanford.nlp.util.Generics;
import edu.stanford.nlp.util.Pair;
import edu.stanford.nlp.util.StringUtils;
/**
* Finds WordNet collocations in parse trees. It can restructure
* collocations as single words, where the original words are joined by
* underscores. You can test performance by using the "collocations" option
* to the TreePrint class.
*
* @author Chris Cox
* @author Eric Yeh
*/
public class CollocationFinder {
private static boolean DEBUG = false;
private final Tree qTree;
private final HeadFinder hf;
private final List collocationCollector;
private final WordNetConnection wnConnect;
/**
* Construct a new {@code CollocationFinder} over the {@code Tree} t.
* The default {@link HeadFinder} is a {@link CollinsHeadFinder}.
* @param t parse tree
* @param w wordnet connection
*/
public CollocationFinder(Tree t, WordNetConnection w) {
this(t, w, new CollinsHeadFinder());
}
/**
* Construct a new {@code CollocationFinder} over the {@code Tree} t.
* @param t parse tree
* @param w wordnet connection
* @param hf {@link HeadFinder} to use
*/
public CollocationFinder(Tree t, WordNetConnection w, HeadFinder hf) {
CoordinationTransformer transformer = new CoordinationTransformer(hf);
this.wnConnect = w;
this.qTree = transformer.transformTree(t);
this.collocationCollector = Generics.newArrayList();
this.hf = hf;
this.getCollocationsList();
if (DEBUG) {
System.err.println("Collected collocations: " + collocationCollector);
}
}
/**
* Returns the "collocations included" parse tree.
*
* @return the mangled tree which applies collocations found in this object.
*/
public Tree getMangledTree() {
return getMangledTree(qTree);
}
private Tree getMangledTree(Tree t) {
Collocation matchingColl = null;
for (Tree child : t.children()) {
child = getMangledTree(child);
}
//boolean additionalCollocationsExist = false;
for (Collocation c : collocationCollector) {
// if there are multiple collocations with the same parent node,
// this will take the longer one
if (t.equals(c.parentNode)) {
if (matchingColl == null ||
(c.span.first() <= matchingColl.span.first() &&
c.span.second() >= matchingColl.span.second())) {
matchingColl = c;
if (DEBUG) {
err.println("Found matching collocation for tree:");
t.pennPrint();
err.print(" head label: " + c.headLabel);
err.println("; collocation string: " + c.collocationString);
err.println(" Constituents: "+ c.indicesOfConstituentChildren);
}
}
}
}
if (matchingColl == null) {
return t;
} else {
if (DEBUG) {
err.println("Collapsing " + matchingColl);
}
Tree[] allChildren = t.children();
// get the earliest child in the collocation and store it as first child.
// delete the rest.
StringBuilder mutatedString = new StringBuilder(160);
for (int i : matchingColl.indicesOfConstituentChildren) {
String strToAppend = mergeLeavesIntoCollocatedString(allChildren[i]);
mutatedString.append(strToAppend);
mutatedString.append("_");
}
mutatedString = mutatedString.deleteCharAt(mutatedString.length() - 1);
// Starting with the latest constituent, delete all the "pruned" children
if (DEBUG) { err.println("allChildren is: " + Arrays.toString(allChildren)); }
for (int index = matchingColl.indicesOfConstituentChildren.size() - 1; index > 0; index--) {
int thisConstituent = matchingColl.indicesOfConstituentChildren.get(index);
allChildren = (Tree[]) ArrayUtils.removeAt(allChildren, thisConstituent);
if (DEBUG) { err.println(" deleted " + thisConstituent + "; allChildren is: " + Arrays.toString(allChildren)); }
}
//name for the leaf string of our new collocation
String newNodeString = mutatedString.toString();
int firstChildIndex = matchingColl.indicesOfConstituentChildren.get(0);
//now we mutate the earliest constituent
Tree newCollocationChild = allChildren[firstChildIndex];
if (DEBUG) err.println("Manipulating: " + newCollocationChild);
newCollocationChild.setValue(matchingColl.headLabel.value());
Tree newCollocationLeaf = newCollocationChild.treeFactory().newLeaf(newNodeString);
newCollocationChild.setChildren(Collections.singletonList(newCollocationLeaf));
if (DEBUG) err.println(" changed to: " + newCollocationChild);
allChildren[firstChildIndex] = newCollocationChild;
t.setChildren(allChildren);
if (DEBUG) {
err.println("Restructured tree is:");
t.pennPrint();
err.println();
}
return t;
}
}
/**
* Traverses the parse tree to find WordNet collocations.
*/
private void getCollocationsList() {
getCollocationsList(qTree);
}
/**
* Prints the collocations found in this Tree as strings.
* Each is followed by its boundary constituent indices in the original tree.
* Example: throw_up (2,3)
* came_up_with (7,9)
*/
public void printCollocationStrings(PrintWriter pw){
//ArrayList strs = new ArrayList();
for (Collocation c: collocationCollector) {
String cs = c.collocationString;
pw.println(cs+" ("+(c.span.first()+1)+","+(c.span.second()+1)+")");
}
}
/**
* This method does the work of traversing the tree and writing collocations
* to the CollocationCollector (an internal data structure).
*
* @param t Tree to get collocations from.
*/
private void getCollocationsList(Tree t) {
int leftMostLeaf = Trees.leftEdge(t,qTree);
if (t.isPreTerminal()) return;
List children = t.getChildrenAsList();
if (children.isEmpty()) return;
//TODO: fix determineHead
// - in phrases like "World Trade Organization 's" the head of the parent NP is "POS".
// - this is problematic for the collocationFinder which assigns this head
// as the POS for the collocation "World_Trade_Organization"!
Label headLabel = hf.determineHead(t).label();
int leftSistersBuffer = 0; //measures the length of sisters in words when reading
for (int i = 0; i < children.size();i++){
ArrayList childConstituents = new ArrayList<>();
childConstituents.add(i);
Tree subtree = children.get(i);
int currWindowLength = 0; //measures the length in words of the current collocation.
getCollocationsList(subtree); //recursive call to get colls in subtrees.
StringBuilder testString = new StringBuilder(160);
testString.append(treeAsStemmedCollocation(subtree));
testString.append('_');
Integer thisSubtreeLength = subtree.yield().size();
currWindowLength+=thisSubtreeLength;
StringBuilder testStringNonStemmed = new StringBuilder(160);
testStringNonStemmed.append(treeAsNonStemmedCollocation(subtree));
testStringNonStemmed.append('_');
//for each subtree i, we iteratively append word yields of succeeding sister
//subtrees j and check their wordnet entries. if they exist we write them to
//the global collocationCollector pair by the indices of the leftmost and
//rightmost words in the collocation.
for (int j = i+1; j < children.size(); j++) {
Tree sisterNode = children.get(j);
childConstituents.add(j);
testString.append(treeAsStemmedCollocation(sisterNode));
testStringNonStemmed.append(treeAsNonStemmedCollocation(sisterNode));
currWindowLength+=sisterNode.yield().size();
if (DEBUG) {
// err.println("Testing string w/ reported indices:" + testString.toString()
// + " (" +(leftMostLeaf+leftSistersBuffer)+","+(leftMostLeaf+leftSistersBuffer+currWindowLength-1)+")");
}
//ignore collocations beginning with "the" or "a"
if (StringUtils.lookingAt(testString.toString(), "(?:[Tt]he|THE|[Aa][Nn]?)[ _]")) {
if (false) {
err.println("CollocationFinder: Not collapsing the/a word: " +
testString);
}
} else if (wordNetContains(testString.toString())) {
Pair c = new Pair<>(leftMostLeaf + leftSistersBuffer, leftMostLeaf + leftSistersBuffer + currWindowLength - 1);
ArrayList childConstituentsClone = new ArrayList<>(childConstituents);
Collocation col = new Collocation(c,t,childConstituentsClone,testString.toString(),headLabel);
collocationCollector.add(col);
if (DEBUG) {
err.println("Found collocation in wordnet: "+ testString);
err.println(" Span of collocation is: " + c +
"; childConstituents is: " + c);
}
}
testString.append('_');
if (StringUtils.lookingAt(testStringNonStemmed.toString(), "(?:[Tt]he|THE|[Aa][Nn]?)[ _]")) {
if (false) {
err.println("CollocationFinder: Not collapsing the/a word: " +
testStringNonStemmed);
}
} else if (wordNetContains(testStringNonStemmed.toString())) {
Pair c = new Pair<>(leftMostLeaf + leftSistersBuffer, leftMostLeaf + leftSistersBuffer + currWindowLength - 1);
ArrayList childConstituentsClone = new ArrayList<>(childConstituents);
Collocation col = new Collocation(c,t,childConstituentsClone,testStringNonStemmed.toString(),headLabel);
collocationCollector.add(col);
if (DEBUG) {
err.println("Found collocation in wordnet: "+ testStringNonStemmed);
err.println(" Span of collocation is: " + c +
"; childConstituents is: " + c);
}
}
testStringNonStemmed.append("_");
}
leftSistersBuffer+=thisSubtreeLength;
}
}
private static String treeAsStemmedCollocation(Tree t) {
List list= getStemmedWordTagsFromTree(t);
// err.println(list.size());
StringBuilder s = new StringBuilder(160);
WordTag firstWord = list.remove(0);
s.append(firstWord.word());
for(WordTag wt : list) {
s.append("_");
s.append(wt.word());
}
//err.println("Expressing this as:"+s.toString());
return s.toString();
}
private static String treeAsNonStemmedCollocation(Tree t) {
List list= getNonStemmedWordTagsFromTree(t);
StringBuilder s = new StringBuilder(160);
WordTag firstWord = list.remove(0);
s.append(firstWord.word());
for(WordTag wt : list) {
s.append('_');
s.append(wt.word());
}
return s.toString();
}
private static String mergeLeavesIntoCollocatedString(Tree t) {
StringBuilder sb = new StringBuilder(160);
ArrayList sent = t.taggedYield();
for (TaggedWord aSent : sent) {
sb.append(aSent.word()).append('_');
}
return sb.substring(0,sb.length() -1);
}
private static String mergeLeavesIntoCollocatedString(Tree[] trees) {
StringBuilder sb = new StringBuilder(160);
for (Tree t: trees) {
ArrayList sent = t.taggedYield();
for (TaggedWord aSent : sent) {
sb.append(aSent.word()).append('_');
}
}
return sb.substring(0,sb.length() -1);
}
/**
*
* @param t a tree
* @return the WordTags corresponding to the leaves of the tree,
* stemmed according to their POS tags in the tree.
*/
private static List getStemmedWordTagsFromTree(Tree t) {
List stemmedWordTags = Generics.newArrayList();
ArrayList s = t.taggedYield();
for (TaggedWord w : s) {
WordTag wt = Morphology.stemStatic(w.word(), w.tag());
stemmedWordTags.add(wt);
}
return stemmedWordTags;
}
private static List getNonStemmedWordTagsFromTree(Tree t) {
List wordTags = Generics.newArrayList();
ArrayList s = t.taggedYield();
for (TaggedWord w : s) {
WordTag wt = new WordTag(w.word(), w.tag());
wordTags.add(wt);
}
return wordTags;
}
/**
* Checks to see if WordNet contains the given word in its lexicon.
* @param s Token
* @return If the given token is in WordNet.
*/
private boolean wordNetContains(String s) {
return wnConnect.wordNetContains(s);
}
/**
* Holds information for one collocation.
*/
private static class Collocation {
final Pair span;
final Tree parentNode;
final Label headLabel;
final List indicesOfConstituentChildren;
final String collocationString;
private Collocation(Pair span,
Tree parentNode,
ArrayList indicesOfConstituentChildren,
String collocationString,
Label headLabel) {
this.span=span;
this.parentNode = parentNode;
this.collocationString=collocationString;
this.indicesOfConstituentChildren=indicesOfConstituentChildren;
this.headLabel=headLabel;
}
@Override
public String toString() {
return collocationString + indicesOfConstituentChildren + "/" +
headLabel;
}
} // end static class Collocation
} // end class CollocationFinder
