Example usage for edu.stanford.nlp.trees Tree numChildren

List of usage examples for edu.stanford.nlp.trees Tree numChildren

  1. HOME
  2. Java
  3. edu.stanford.nlp
  4. edu.stanford.nlp.trees.*
  5. Tree
  6. numChildren

Introduction

In this page you can find the example usage for edu.stanford.nlp.trees Tree numChildren.

Prototype

public int numChildren()

Source Link

Document

Says how many children a tree node has in its local tree.

Usage

From source file:KleinBilingualParser.java

private static double numChildren(Tree nodeF, Tree nodeE) {
    if (nodeF.numChildren() == nodeE.numChildren()) {
        return 1.0 / 10;
    } else {/*from   w ww . ja v a 2 s. com*/
        return 0.0;
    }
}

From source file:conditionalCFG.ConditionalCFGParser.java

License:Open Source License

public double validateBinarizedTree(Tree tree, int start) {
    if (tree.isLeaf()) {
        return 0.0;
    }//from   w  ww . ja  va 2 s.  c o m
    float epsilon = 0.0001f;
    if (tree.isPreTerminal()) {
        String wordStr = tree.children()[0].label().value();
        int tag = tagIndex.indexOf(tree.label().value());
        int word = wordIndex.indexOf(wordStr);
        IntTaggedWord iTW = new IntTaggedWord(word, tag);
        float score = lex.score(iTW, start, wordStr, null);
        float bound = iScore[start][start + 1][stateIndex.indexOf(tree.label().value())];
        if (score > bound + epsilon) {
            System.out.println("Invalid tagging:");
            System.out.println("  Tag: " + tree.label().value());
            System.out.println("  Word: " + tree.children()[0].label().value());
            System.out.println("  Score: " + score);
            System.out.println("  Bound: " + bound);
        }
        return score;
    }
    int parent = stateIndex.indexOf(tree.label().value());
    int firstChild = stateIndex.indexOf(tree.children()[0].label().value());
    if (tree.numChildren() == 1) {
        UnaryRule ur = new UnaryRule(parent, firstChild);
        double score = SloppyMath.max(ug.scoreRule(ur), -10000.0)
                + validateBinarizedTree(tree.children()[0], start);
        double bound = iScore[start][start + tree.yield().size()][parent];
        if (score > bound + epsilon) {
            System.out.println("Invalid unary:");
            System.out.println("  Parent: " + tree.label().value());
            System.out.println("  Child: " + tree.children()[0].label().value());
            System.out.println("  Start: " + start);
            System.out.println("  End: " + (start + tree.yield().size()));
            System.out.println("  Score: " + score);
            System.out.println("  Bound: " + bound);
        }
        return score;
    }
    int secondChild = stateIndex.indexOf(tree.children()[1].label().value());
    BinaryRule br = new BinaryRule(parent, firstChild, secondChild);
    double score = SloppyMath.max(bg.scoreRule(br), -10000.0) + validateBinarizedTree(tree.children()[0], start)
            + validateBinarizedTree(tree.children()[1], start + tree.children()[0].yield().size());
    double bound = iScore[start][start + tree.yield().size()][parent];
    if (score > bound + epsilon) {
        System.out.println("Invalid binary:");
        System.out.println("  Parent: " + tree.label().value());
        System.out.println("  LChild: " + tree.children()[0].label().value());
        System.out.println("  RChild: " + tree.children()[1].label().value());
        System.out.println("  Start: " + start);
        System.out.println("  End: " + (start + tree.yield().size()));
        System.out.println("  Score: " + score);
        System.out.println("  Bound: " + bound);
    }
    return score;
}

From source file:conditionalCFG.ConditionalCFGParser.java

License:Open Source License

public double scoreBinarizedTree(Tree tree, int start, int debugLvl) {
    if (tree.isLeaf()) {
        return 0.0;
    }//from   www.  ja v a2s .c om
    if (tree.isPreTerminal()) {
        String wordStr = tree.children()[0].label().value();
        int tag = tagIndex.indexOf(tree.label().value());
        int word = wordIndex.indexOf(wordStr);
        IntTaggedWord iTW = new IntTaggedWord(word, tag);
        // if (lex.score(iTW,(leftmost ? 0 : 1)) == Double.NEGATIVE_INFINITY) {
        //   System.out.println("NO SCORE FOR: "+iTW);
        // }
        float score = lex.score(iTW, start, wordStr, null);
        tree.setScore(score);
        if (debugLvl > 0)
            System.out.println(score + " " + tree.getSpan());
        return score;
    }
    int parent = stateIndex.indexOf(tree.label().value());
    int firstChild = stateIndex.indexOf(tree.children()[0].label().value());
    if (tree.numChildren() == 1) {
        UnaryRule ur = new UnaryRule(parent, firstChild);
        //+ DEBUG
        // if (ug.scoreRule(ur) < -10000) {
        //        System.out.println("Grammar doesn't have rule: " + ur);
        // }
        //      return SloppyMath.max(ug.scoreRule(ur), -10000.0) + scoreBinarizedTree(tree.children()[0], leftmost);
        double score = ug.scoreRule(ur) + scoreBinarizedTree(tree.children()[0], start, debugLvl)
                + lex.score(ur, start, start + tree.children()[0].yield().size());
        tree.setScore(score);
        if (debugLvl > 0)
            System.out.println(score + " " + tree.getSpan());
        return score;
    }
    int secondChild = stateIndex.indexOf(tree.children()[1].label().value());
    BinaryRule br = new BinaryRule(parent, firstChild, secondChild);
    //+ DEBUG
    // if (bg.scoreRule(br) < -10000) {
    //  System.out.println("Grammar doesn't have rule: " + br);
    // }
    //    return SloppyMath.max(bg.scoreRule(br), -10000.0) +
    //            scoreBinarizedTree(tree.children()[0], leftmost) +
    //            scoreBinarizedTree(tree.children()[1], false);
    int sz0 = tree.children()[0].yield().size();
    double score = bg.scoreRule(br) + scoreBinarizedTree(tree.children()[0], start, debugLvl)
            + scoreBinarizedTree(tree.children()[1], start + sz0, debugLvl)
            + lex.score(br, start, start + sz0 + tree.children()[1].yield().size(), start + sz0);
    tree.setScore(score);
    if (debugLvl > 0)
        System.out.println(score + " " + tree.getSpan() + " " + (sz0 + start));
    return score;
}

From source file:edu.cmu.ark.AnalysisUtilities.java

License:Open Source License

public static boolean cCommands(Tree root, Tree n1, Tree n2) {
    if (n1.dominates(n2))
        return false;

    Tree n1Parent = n1.parent(root);
    while (n1Parent != null && n1Parent.numChildren() == 1) {
        n1Parent = n1Parent.parent(root);
    }//w w  w . jav a 2s.  co m

    if (n1Parent != null && n1Parent.dominates(n2))
        return true;

    return false;
}

From source file:edu.cmu.ark.nlp.sent.SentenceSimplifier.java

License:Open Source License

/**
 * e.g., John and James like Susan.  ->  John likes Susan.
 * //www .j  av  a2 s.com
 */
private void extractConjoinedNPs(Collection<Question> extracted, Question input) {
    String tregexOpStr;
    TregexPattern matchPattern;
    Tree conjoinedNode;
    Tree parent;

    TregexMatcher matcher;
    Question newQuestion;

    //only extract conjoined NPs that are arguments or adjuncts of the main verb
    // in the tree, this means the closest S will be the one under the root
    tregexOpStr = "NP=parent < (CONJP|CC !< or|nor [ "
            + " $+ /^(N.*|PRP|SBAR)$/=child $-- /^(N.*|PRP|SBAR)$/ | " //there must be a noun on each side of the conjunction
            + " $-- /^(N.*|PRP|SBAR)$/=child $+ /^(N.*|PRP|SBAR)$/ ] ) " //this avoids extracting from flat NPs such as "the smaller and darker form"
            + " !>> (/.*/ $ (CC|CONJP !< or|nor)) " //this cannot be nested within a larger conjunction or followed by a conjunction (we recur later to catch this) 
            + " !$ (CC|CONJP !< or|nor)" + " !.. (CC|CONJP !< or|nor > NP|PP|S|SBAR|VP) !>> SBAR ";
    //+ " >> (ROOT !< (S <+(VP) (/^VB.*$/ < are|were|be|seem|appear))) " ; //don't break plural predicate nominatives (e.g., "John and Mary are two of my best friends.")
    matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
    matcher = matchPattern.matcher(input.getIntermediateTree());
    List<Integer> nodeIndexes = new ArrayList<Integer>();
    List<Integer> parentIDs = new ArrayList<Integer>();

    while (matcher.find()) {
        //store the parents' IDs (in the tree)
        parent = matcher.getNode("parent");
        parentIDs.add(parent.nodeNumber(input.getIntermediateTree()));

        conjoinedNode = matcher.getNode("child");
        //store the conjoined nodes' index into their parent's list of children
        int idx = parent.objectIndexOf(conjoinedNode);
        if (!nodeIndexes.contains(idx))
            nodeIndexes.add(idx);
    }

    //for each of the conjoined children,
    //create a new tree by removing all the nodes they are conjoined with
    Collections.sort(nodeIndexes);//sort, just to keep them in the original order
    for (int i = 0; i < nodeIndexes.size(); i++) {
        newQuestion = input.deeperCopy();

        Tree t = newQuestion.getIntermediateTree();
        parent = t.getNodeNumber(parentIDs.get(i));
        Tree gparent = parent.parent(t);
        conjoinedNode = parent.getChild(nodeIndexes.get(i));
        String siblingLabel;

        //Remove all the nodes that are conjoined
        //with the selected noun (or are conjunctions, commas).
        //These can have labels NP, NN, ..., PRP for pronouns, CC, "," for commas, ":" for semi-colons
        for (int j = 0; j < parent.numChildren(); j++) {
            if (parent.getChild(j) == conjoinedNode)
                continue;
            siblingLabel = parent.getChild(j).label().toString();
            if (siblingLabel.matches("^[NCP,:S].*")) {
                parent.removeChild(j);
                j--;
            }
        }

        //if there is an trivial unary "NP -> NP",
        //remove the parent and put the child in its place
        if (parent.numChildren() == 1 && parent.getChild(0).label().equals("NP")) {
            int tmpIndex = gparent.objectIndexOf(parent);
            gparent.removeChild(tmpIndex);
            gparent.addChild(tmpIndex, parent.getChild(0));
        }

        correctTense(conjoinedNode, gparent);
        addQuotationMarksIfNeeded(newQuestion.getIntermediateTree());

        //if(GlobalProperties.getDebug()) System.err.println("extractConjoinedNPs: "+newQuestion.getIntermediateTree().toString());
        if (this.getComputeFeatures)
            newQuestion.setFeatureValue("extractedFromConjoinedPhrases", 1.0); //old feature name
        if (this.getComputeFeatures)
            newQuestion.setFeatureValue("extractedFromConjoinedNPs", 1.0);
        extracted.add(newQuestion);
    }
}

From source file:edu.cmu.ark.nlp.sent.SentenceSimplifier.java

License:Open Source License

/**
 * e.g., John, who is a friend of mine, likes Susan. -> John is a friend of mine.
 * /*from w  w w . j  a  v  a 2s  .  co m*/
 */
private void extractNonRestrictiveRelativeClauses(Collection<Question> extracted, Question input) {
    String tregexOpStr;
    TregexPattern matchPattern;
    TregexMatcher matcher;
    TregexMatcher matcherclause;

    tregexOpStr = "NP=np < (SBAR=sbar [ < (WHADVP=wherecomp < (WRB < where)) " + " | < (WHNP !< /WP\\$/) "
            + " | < (WHNP=possessive < /WP\\$/)" //John, whose car was
            + " | < (WHPP < IN|TO=preposition) ] $-- NP $- /,/ " + " < S=relclause  !< WHADJP)";

    matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
    matcher = matchPattern.matcher(input.getIntermediateTree());

    //iterate over all the relative clauses in the input
    //and create an output sentence for each one.
    while (matcher.find()) {
        Tree missingArgumentTree = matcher.getNode("np");
        Tree relclause = matcher.getNode("relclause");
        if (missingArgumentTree == null || relclause == null)
            continue;
        missingArgumentTree = missingArgumentTree.deepCopy();
        relclause = relclause.deepCopy();
        Tree possessive = matcher.getNode("possessive");
        Tree sbar = matcher.getNode("sbar").deepCopy();

        makeDeterminerDefinite(missingArgumentTree);

        if (possessive != null) {
            possessive = possessive.deepCopy();
            possessive.removeChild(0);
            String newTree = "(NP (NP " + missingArgumentTree.toString() + " (POS 's))";
            for (int i = 0; i < possessive.numChildren(); i++)
                newTree += possessive.getChild(i).toString() + " ";
            newTree += ")";
            missingArgumentTree = QuestionUtil.readTreeFromString(newTree);
        }

        //remove the relative clause and the commas surrounding it from the missing argument tree
        for (int i = 0; i < missingArgumentTree.numChildren(); i++) {
            if (missingArgumentTree.getChild(i).equals(sbar)) {
                //remove the relative clause
                missingArgumentTree.removeChild(i);
                //remove the comma after the relative clause
                if (i < missingArgumentTree.numChildren()
                        && missingArgumentTree.getChild(i).label().toString().equals(",")) {
                    missingArgumentTree.removeChild(i);
                }
                //remove the comma before the relative clause
                if (i > 0 && missingArgumentTree.getChild(i - 1).label().toString().equals(",")) {
                    missingArgumentTree.removeChild(i - 1);
                    i--;
                }
                i--;
            }
        }

        //put the noun in the clause at the topmost place with an opening for a noun. 
        //Note that this may mess up if there are noun phrase adjuncts like "The man I met Tuesday".

        //specifically: 
        //the parent of the noun can be either a clause (S) as in "The man who met me"
        //or a verb phrase as in "The man who I met".
        //for verb phrases, add the noun to the end since it will be an object.
        //for clauses, add the noun to the beginning since it will be the subject.
        tregexOpStr = "S|VP=newparent !< NP < (VP=verb !< TO !$ TO)";
        matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
        matcherclause = matchPattern.matcher(relclause);
        boolean subjectMovement = true;
        if (!matcherclause.find()) {
            tregexOpStr = "VP=newparent !< VP < /VB.*/=verb !>> (S !< NP) !<< (VP !< VP !< NP)";
            matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
            matcherclause = matchPattern.matcher(relclause);
            subjectMovement = false;
        }

        //reset (so the first match isn't skipped)
        matcherclause = matchPattern.matcher(relclause);

        if (matcherclause.find()) {
            Tree newparenttree = matcherclause.getNode("newparent");
            Tree verbtree = matcherclause.getNode("verb");
            boolean ppRelativeClause = false;

            if (matcher.getNode("wherecomp") != null) {
                String tmp = "(PP (IN at) " + missingArgumentTree.toString() + ")";
                missingArgumentTree = QuestionUtil.readTreeFromString(tmp);
                ppRelativeClause = true;
                subjectMovement = false;
            } else if (matcher.getNode("preposition") != null) {
                String tmp = "(PP (IN " + matcher.getNode("preposition").yield().toString() + ") "
                        + missingArgumentTree.toString() + ")";
                missingArgumentTree = QuestionUtil.readTreeFromString(tmp);
                ppRelativeClause = true;
            }

            if (subjectMovement) { //subject
                newparenttree.addChild(newparenttree.objectIndexOf(verbtree), missingArgumentTree);
            } else { // newparentlabel is VP   
                if (ppRelativeClause)
                    newparenttree.addChild(newparenttree.numChildren(), missingArgumentTree);
                else
                    newparenttree.addChild(newparenttree.objectIndexOf(verbtree) + 1, missingArgumentTree);
            }

            //create a new tree with punctuation
            Tree newTree = factory.newTreeNode("ROOT", new ArrayList<Tree>());
            newTree.addChild(relclause);
            QuestionUtil.addPeriodIfNeeded(newTree);

            //if(GlobalProperties.getDebug()) System.err.println("extractRelativeClauses: "+ newTree.toString());
            addQuotationMarksIfNeeded(newTree);
            Question newTreeWithFeatures = input.deeperCopy();
            newTreeWithFeatures.setIntermediateTree(newTree);
            if (this.getComputeFeatures)
                newTreeWithFeatures.setFeatureValue("extractedFromRelativeClause", 1.0);
            addIfNovel(extracted, newTreeWithFeatures);
        }
    }
}

From source file:edu.cmu.ark.QuestionTransducer.java

License:Open Source License

/**
 *
 * This method removes the answer phrase from its original position
 * and places it at the front of the main clause.
 *
 * Note: Tsurgeon operations are perhaps not optimal here.
 * Using the Stanford API to move nodes directly might be simpler...
 *
 */// w  w  w . jav a  2s  .com
private List<Tree> moveWHPhraseUnmovable(Tree inputTree, Tree intermediateTree, int i,
        boolean subjectMovement) {
    Tree copyTree2;
    List<Tree> res = new ArrayList<Tree>();
    Tree mainclauseNode;

    String marker = "/^(UNMOVABLE-NP|UNMOVABLE-PP|UNMOVABLE-SBAR)-" + i + "$/";

    List<Pair<TregexPattern, TsurgeonPattern>> ops = new ArrayList<Pair<TregexPattern, TsurgeonPattern>>();
    List<TsurgeonPattern> ps = new ArrayList<TsurgeonPattern>();
    String tregexOpStr;
    TregexPattern matchPattern;

    //extract the "answer" phrase and generate a WH phrase from it
    tregexOpStr = "ROOT=root < (SQ=qclause << " + marker + "=answer < VP=predicate)";
    matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
    if (GlobalProperties.getDebug())
        System.err.println("moveWHPhrase: inputTree:" + inputTree.toString());
    if (GlobalProperties.getDebug())
        System.err.println("moveWHPhrase: tregexOpStr:" + tregexOpStr);
    TregexMatcher matcher = matchPattern.matcher(inputTree);
    matcher.find();
    Tree phraseToMove = matcher.getNode("answer");

    String whPhraseSubtree;

    if (printExtractedPhrases)
        System.out.println("EXTRACTED\t" + phraseToMove.yield().toString());

    whGen.generateWHPhraseSubtrees(removeMarkersFromTree(phraseToMove), intermediateTree.yield().toString());
    List<String> whPhraseSubtrees = whGen.getWHPhraseSubtrees();
    List<String> leftOverPrepositions = whGen.getLeftOverPrepositions();

    //copyTree = inputTree.deeperCopy();
    //The placeholder is necessary because tsurgeon will complain
    //if an added node has no children. This placeholder is removed below.
    //      ps.add(Tsurgeon.parseOperation("insert (PREPPLACEHOLDER dummy) $+ answer"));
    //      ps.add(Tsurgeon.parseOperation("prune answer"));
    //      ps.add(Tsurgeon.parseOperation("insert (SBARQ=mainclause PLACEHOLDER=placeholder) >0 root"));
    //      ps.add(Tsurgeon.parseOperation("move qclause >-1 mainclause"));
    //      p = Tsurgeon.collectOperations(ps);
    //      ops.add(new Pair<TregexPattern,TsurgeonPattern>(matchPattern,p));
    //      Tsurgeon.processPatternsOnTree(ops, copyTree);

    //copyTree = removeMarkersFromTree(copyTree);

    //Now put each WH phrase into the tree and remove the original answer.
    //Operate on the tree directly rather than using tsurgeon
    //because tsurgeon can't parse operations that insert trees with special characters (e.g., ":")
    for (int j = 0; j < whPhraseSubtrees.size(); j++) {
        copyTree2 = inputTree.deeperCopy();
        whPhraseSubtree = whPhraseSubtrees.get(j);

        //         if(GlobalProperties.getDebug()) System.err.println("moveWHPhrase: whPhraseSubtree:"+whPhraseSubtree);
        //         tregexOpStr = "ROOT < (SBARQ=mainclause < PLACEHOLDER=ph1) << (__=ph2Parent < PREPPLACEHOLDER=ph2)";
        //         matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
        //         matcher = matchPattern.matcher(copyTree2);
        //         if(!matcher.find()){
        //            continue;
        //         }
        matcher = matchPattern.matcher(copyTree2);
        matcher.find();
        mainclauseNode = matcher.getNode("answer");
        if (mainclauseNode == null)
            continue;
        //replace the wh placeholder with a wh phrase
        int cc = mainclauseNode.numChildren();
        for (int c = 0; c < cc; c++)
            mainclauseNode.removeChild(0);
        mainclauseNode.addChild(0, AnalysisUtilities.getInstance().readTreeFromString(whPhraseSubtree));

        copyTree2 = removeMarkersFromTree(copyTree2);
        //Replace the pp placeholder with the left over preposition.
        //This may happen when the answer phrase was a PP.
        //e.g., John went to the game. -> What did John go to?
        //         prepPlaceholderParent = matcher.getNode("ph2Parent");
        //         int index = prepPlaceholderParent.indexOf(matcher.getNode("ph2"));
        //         if(leftOverPreposition != null && leftOverPreposition.length()>0){
        //            prepPlaceholderParent.addChild(index, AnalysisUtilities.getInstance().readTreeFromString(leftOverPreposition));
        //         }
        //         //now remove the left-over-preposition placeholder
        //         ps.clear();
        //         ps.add(Tsurgeon.parseOperation("prune ph2"));
        //         p = Tsurgeon.collectOperations(ps);
        //         ops.clear();
        //         ops.add(new Pair<TregexPattern,TsurgeonPattern>(TregexPatternFactory.getPattern("PREPPLACEHOLDER=ph2"),p));
        //         Tsurgeon.processPatternsOnTree(ops, copyTree2);

        copyTree2 = moveLeadingAdjuncts(copyTree2);

        if (GlobalProperties.getDebug())
            System.err.println("moveWHPhrase: " + copyTree2.toString());
        res.add(copyTree2);
    }

    return res;
}

From source file:edu.cmu.ark.SentenceSimplifier.java

License:Open Source License

/**
 * e.g., John and James like Susan.  ->  John likes Susan.
 * /* w ww .  java 2s .c o  m*/
 */
private void extractConjoinedNPs(Collection<Question> extracted, Question input) {
    String tregexOpStr;
    TregexPattern matchPattern;
    Tree conjoinedNode;
    Tree parent;

    TregexMatcher matcher;
    Question newQuestion;

    //only extract conjoined NPs that are arguments or adjuncts of the main verb
    // in the tree, this means the closest S will be the one under the root
    tregexOpStr = "NP=parent < (CONJP|CC !< or|nor [ "
            + " $+ /^(N.*|PRP|SBAR)$/=child $-- /^(N.*|PRP|SBAR)$/ | " //there must be a noun on each side of the conjunction
            + " $-- /^(N.*|PRP|SBAR)$/=child $+ /^(N.*|PRP|SBAR)$/ ] ) " //this avoids extracting from flat NPs such as "the smaller and darker form"
            + " !>> (/.*/ $ (CC|CONJP !< or|nor)) " //this cannot be nested within a larger conjunction or followed by a conjunction (we recur later to catch this) 
            + " !$ (CC|CONJP !< or|nor)" + " !.. (CC|CONJP !< or|nor > NP|PP|S|SBAR|VP) !>> SBAR ";
    //+ " >> (ROOT !< (S <+(VP) (/^VB.*$/ < are|were|be|seem|appear))) " ; //don't break plural predicate nominatives (e.g., "John and Mary are two of my best friends.")
    matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
    matcher = matchPattern.matcher(input.getIntermediateTree());
    List<Integer> nodeIndexes = new ArrayList<Integer>();
    List<Integer> parentIDs = new ArrayList<Integer>();

    while (matcher.find()) {
        //store the parents' IDs (in the tree)
        parent = matcher.getNode("parent");
        parentIDs.add(parent.nodeNumber(input.getIntermediateTree()));

        conjoinedNode = matcher.getNode("child");
        //store the conjoined nodes' index into their parent's list of children
        int idx = parent.indexOf(conjoinedNode);
        if (!nodeIndexes.contains(idx))
            nodeIndexes.add(idx);
    }

    //for each of the conjoined children,
    //create a new tree by removing all the nodes they are conjoined with
    Collections.sort(nodeIndexes);//sort, just to keep them in the original order
    for (int i = 0; i < nodeIndexes.size(); i++) {
        newQuestion = input.deeperCopy();

        Tree t = newQuestion.getIntermediateTree();
        parent = t.getNodeNumber(parentIDs.get(i));
        Tree gparent = parent.parent(t);
        conjoinedNode = parent.getChild(nodeIndexes.get(i));
        String siblingLabel;

        //Remove all the nodes that are conjoined
        //with the selected noun (or are conjunctions, commas).
        //These can have labels NP, NN, ..., PRP for pronouns, CC, "," for commas, ":" for semi-colons
        for (int j = 0; j < parent.numChildren(); j++) {
            if (parent.getChild(j) == conjoinedNode)
                continue;
            siblingLabel = parent.getChild(j).label().toString();
            if (siblingLabel.matches("^[NCP,:S].*")) {
                parent.removeChild(j);
                j--;
            }
        }

        //if there is an trivial unary "NP -> NP",
        //remove the parent and put the child in its place
        if (parent.numChildren() == 1 && parent.getChild(0).label().equals("NP")) {
            int tmpIndex = gparent.indexOf(parent);
            gparent.removeChild(tmpIndex);
            gparent.addChild(tmpIndex, parent.getChild(0));
        }

        correctTense(conjoinedNode, gparent);
        addQuotationMarksIfNeeded(newQuestion.getIntermediateTree());

        if (GlobalProperties.getDebug())
            System.err.println("extractConjoinedNPs: " + newQuestion.getIntermediateTree().toString());
        if (GlobalProperties.getComputeFeatures())
            newQuestion.setFeatureValue("extractedFromConjoinedPhrases", 1.0); //old feature name
        if (GlobalProperties.getComputeFeatures())
            newQuestion.setFeatureValue("extractedFromConjoinedNPs", 1.0);
        extracted.add(newQuestion);
    }
}

From source file:edu.cmu.ark.SentenceSimplifier.java

License:Open Source License

/**
 * e.g., John, who is a friend of mine, likes Susan. -> John is a friend of mine.
 * //from w  w  w  . jav a  2 s .  c o  m
 */
private void extractNonRestrictiveRelativeClauses(Collection<Question> extracted, Question input) {
    String tregexOpStr;
    TregexPattern matchPattern;
    TregexMatcher matcher;
    TregexMatcher matcherclause;

    tregexOpStr = "NP=np < (SBAR=sbar [ < (WHADVP=wherecomp < (WRB < where)) " + " | < (WHNP !< /WP\\$/) "
            + " | < (WHNP=possessive < /WP\\$/)" //John, whose car was
            + " | < (WHPP < IN|TO=preposition) ] $-- NP $- /,/ " + " < S=relclause  !< WHADJP)";

    matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
    matcher = matchPattern.matcher(input.getIntermediateTree());

    //iterate over all the relative clauses in the input
    //and create an output sentence for each one.
    while (matcher.find()) {
        Tree missingArgumentTree = matcher.getNode("np");
        Tree relclause = matcher.getNode("relclause");
        if (missingArgumentTree == null || relclause == null)
            continue;
        missingArgumentTree = missingArgumentTree.deeperCopy();
        relclause = relclause.deeperCopy();
        Tree possessive = matcher.getNode("possessive");
        Tree sbar = matcher.getNode("sbar").deeperCopy();

        makeDeterminerDefinite(missingArgumentTree);

        if (possessive != null) {
            possessive = possessive.deeperCopy();
            possessive.removeChild(0);
            String newTree = "(NP (NP " + missingArgumentTree.toString() + " (POS 's))";
            for (int i = 0; i < possessive.numChildren(); i++)
                newTree += possessive.getChild(i).toString() + " ";
            newTree += ")";
            missingArgumentTree = AnalysisUtilities.getInstance().readTreeFromString(newTree);
        }

        //remove the relative clause and the commas surrounding it from the missing argument tree
        for (int i = 0; i < missingArgumentTree.numChildren(); i++) {
            if (missingArgumentTree.getChild(i).equals(sbar)) {
                //remove the relative clause
                missingArgumentTree.removeChild(i);
                //remove the comma after the relative clause
                if (i < missingArgumentTree.numChildren()
                        && missingArgumentTree.getChild(i).label().toString().equals(",")) {
                    missingArgumentTree.removeChild(i);
                }
                //remove the comma before the relative clause
                if (i > 0 && missingArgumentTree.getChild(i - 1).label().toString().equals(",")) {
                    missingArgumentTree.removeChild(i - 1);
                    i--;
                }
                i--;
            }
        }

        //put the noun in the clause at the topmost place with an opening for a noun. 
        //Note that this may mess up if there are noun phrase adjuncts like "The man I met Tuesday".

        //specifically: 
        //the parent of the noun can be either a clause (S) as in "The man who met me"
        //or a verb phrase as in "The man who I met".
        //for verb phrases, add the noun to the end since it will be an object.
        //for clauses, add the noun to the beginning since it will be the subject.
        tregexOpStr = "S|VP=newparent !< NP < (VP=verb !< TO !$ TO)";
        matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
        matcherclause = matchPattern.matcher(relclause);
        boolean subjectMovement = true;
        if (!matcherclause.find()) {
            tregexOpStr = "VP=newparent !< VP < /VB.*/=verb !>> (S !< NP) !<< (VP !< VP !< NP)";
            matchPattern = TregexPatternFactory.getPattern(tregexOpStr);
            matcherclause = matchPattern.matcher(relclause);
            subjectMovement = false;
        }

        //reset (so the first match isn't skipped)
        matcherclause = matchPattern.matcher(relclause);

        if (matcherclause.find()) {
            Tree newparenttree = matcherclause.getNode("newparent");
            Tree verbtree = matcherclause.getNode("verb");
            boolean ppRelativeClause = false;

            if (matcher.getNode("wherecomp") != null) {
                String tmp = "(PP (IN at) " + missingArgumentTree.toString() + ")";
                missingArgumentTree = AnalysisUtilities.getInstance().readTreeFromString(tmp);
                ppRelativeClause = true;
                subjectMovement = false;
            } else if (matcher.getNode("preposition") != null) {
                String tmp = "(PP (IN " + matcher.getNode("preposition").yield().toString() + ") "
                        + missingArgumentTree.toString() + ")";
                missingArgumentTree = AnalysisUtilities.getInstance().readTreeFromString(tmp);
                ppRelativeClause = true;
            }

            if (subjectMovement) { //subject
                newparenttree.addChild(newparenttree.indexOf(verbtree), missingArgumentTree);
            } else { // newparentlabel is VP   
                if (ppRelativeClause)
                    newparenttree.addChild(newparenttree.numChildren(), missingArgumentTree);
                else
                    newparenttree.addChild(newparenttree.indexOf(verbtree) + 1, missingArgumentTree);
            }

            //create a new tree with punctuation
            Tree newTree = factory.newTreeNode("ROOT", new ArrayList<Tree>());
            newTree.addChild(relclause);
            AnalysisUtilities.addPeriodIfNeeded(newTree);

            if (GlobalProperties.getDebug())
                System.err.println("extractRelativeClauses: " + newTree.toString());
            addQuotationMarksIfNeeded(newTree);
            Question newTreeWithFeatures = input.deeperCopy();
            newTreeWithFeatures.setIntermediateTree(newTree);
            if (GlobalProperties.getComputeFeatures())
                newTreeWithFeatures.setFeatureValue("extractedFromRelativeClause", 1.0);
            addIfNovel(extracted, newTreeWithFeatures);
        }
    }
}

From source file:edu.jhu.hlt.concrete.stanford.PreNERCoreMapWrapper.java

License:Open Source License

/**
*
* @param root//from ww  w  .j  av  a 2s. co  m
* @param left
* @param right
* @param n
*          is the length of the sentence is tokens.
* @param p
* @param tokenizationUUID
* @return The constituent ID
* @throws AnalyticException
*/
private static int constructConstituent(Tree root, int left, int right, int n, Parse p, UUID tokenizationUUID,
        HeadFinder hf) throws AnalyticException {

    Constituent constituent = new Constituent();
    constituent.setId(p.getConstituentListSize());
    constituent.setTag(root.value());
    constituent.setStart(left);
    constituent.setEnding(right);
    p.addToConstituentList(constituent);
    Tree headTree = null;
    if (!root.isLeaf()) {
        try {
            headTree = hf.determineHead(root);
        } catch (java.lang.IllegalArgumentException iae) {
            LOGGER.warn("Failed to find head, falling back on rightmost constituent.");
            headTree = root.children()[root.numChildren() - 1];
        }
    }
    int i = 0, headTreeIdx = -1;

    int leftPtr = left;
    for (Tree child : root.getChildrenAsList()) {
        int width = child.getLeaves().size();
        int childId = constructConstituent(child, leftPtr, leftPtr + width, n, p, tokenizationUUID, hf);
        constituent.addToChildList(childId);

        leftPtr += width;
        if (headTree != null && child == headTree) {
            assert (headTreeIdx < 0);
            headTreeIdx = i;
        }
        i++;
    }

    if (headTreeIdx >= 0)
        constituent.setHeadChildIndex(headTreeIdx);

    if (!constituent.isSetChildList())
        constituent.setChildList(new ArrayList<Integer>());
    return constituent.getId();
}