Examples with IndexedWord edu.stanford.nlp.ling.IndexedWord used on opensource projects

Example 6 with IndexedWord

use of edu.stanford.nlp.ling.IndexedWord in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method expandPPConjunctions.

/**
   * Expands PPs with conjunctions such as in the sentence
   * "Bill flies to France and from Serbia." by copying the verb
   * that governs the prepositinal phrase resulting in the following
   * relations:
   * <p/>
   * {@code conj:and(flies, flies')}<br/>
   * {@code case(France, to)}<br/>
   * {@code cc(flies, and)}<br/>
   * {@code case(Serbia, from)}<br/>
   * {@code nmod(flies, France)}<br/>
   * {@code nmod(flies', Serbia)}<br/>
   * <p/>
   * The label of the conjunct relation includes the conjunction type
   * because if the verb has multiple cc relations then it can be impossible
   * to infer which coordination marker belongs to which conjuncts.
   *
   * @param sg SemanticGraph to operate on.
   */
private static void expandPPConjunctions(SemanticGraph sg) {
    /* Semgrexes require a graph with a root. */
    if (sg.getRoots().isEmpty())
        return;
    SemanticGraph sgCopy = sg.makeSoftCopy();
    SemgrexMatcher matcher = PP_CONJP_PATTERN.matcher(sgCopy);
    IndexedWord oldGov = null;
    IndexedWord oldCcDep = null;
    List<IndexedWord> conjDeps = Generics.newLinkedList();
    while (matcher.find()) {
        IndexedWord conjDep = matcher.getNode("conj");
        IndexedWord gov = matcher.getNode("gov");
        IndexedWord ccDep = matcher.getNode("cc");
        if (oldGov != null && (!gov.equals(oldGov) || !ccDep.equals(oldCcDep))) {
            expandPPConjunction(sg, oldGov, conjDeps, oldCcDep);
            conjDeps = Generics.newLinkedList();
        }
        oldCcDep = ccDep;
        oldGov = gov;
        conjDeps.add(conjDep);
    }
    if (oldGov != null) {
        expandPPConjunction(sg, oldGov, conjDeps, oldCcDep);
    }
}

Example 7 with IndexedWord

use of edu.stanford.nlp.ling.IndexedWord in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method correctWHAttachment.

/**
   * Tries to correct complicated cases of WH-movement in
   * sentences such as "What does Mary seem to have?" in
   * which "What" should attach to "have" instead of the
   * control verb.
   *
   * @param sg The Semantic graph to operate on.
   */
private static void correctWHAttachment(SemanticGraph sg) {
    /* Semgrexes require a graph with a root. */
    if (sg.getRoots().isEmpty())
        return;
    SemanticGraph sgCopy = sg.makeSoftCopy();
    SemgrexMatcher matcher = XCOMP_PATTERN.matcher(sgCopy);
    while (matcher.findNextMatchingNode()) {
        IndexedWord root = matcher.getNode("root");
        IndexedWord embeddedVerb = matcher.getNode("embedded");
        IndexedWord wh = matcher.getNode("wh");
        IndexedWord dobj = matcher.getNode("obj");
        /* Check if the object is a WH-word. */
        if (wh.tag().startsWith("W")) {
            boolean reattach = false;
            /* If the control verb already has an object, then
           we have to reattach the WH-word to the verb in the embedded clause. */
            if (dobj != null) {
                reattach = true;
            } else {
                /* If the control verb can't have an object, we also have to reattach. */
                String lemma = Morphology.lemmaStatic(root.value(), root.tag());
                if (lemma.matches(EnglishPatterns.NP_V_S_INF_VERBS_REGEX)) {
                    reattach = true;
                }
            }
            if (reattach) {
                SemanticGraphEdge edge = sg.getEdge(root, wh);
                if (edge != null) {
                    sg.removeEdge(edge);
                    sg.addEdge(embeddedVerb, wh, DIRECT_OBJECT, Double.NEGATIVE_INFINITY, false);
                }
            }
        }
    }
}

Example 8 with IndexedWord

use of edu.stanford.nlp.ling.IndexedWord in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method processComplex2WP.

/**
   * Processes all the two-word prepositions in TWO_WORD_PREPS_COMPLEX.
   */
private static void processComplex2WP(SemanticGraph sg, HashMap<String, HashSet<Integer>> bigrams) {
    for (String bigram : TWO_WORD_PREPS_COMPLEX) {
        if (bigrams.get(bigram) == null) {
            continue;
        }
        for (Integer i : bigrams.get(bigram)) {
            IndexedWord w1 = sg.getNodeByIndexSafe(i);
            IndexedWord w2 = sg.getNodeByIndexSafe(i + 1);
            if (w1 == null || w2 == null) {
                continue;
            }
            SemgrexMatcher matcher = TWO_WORD_PREPS_COMPLEX_PATTERN.matcher(sg);
            IndexedWord gov = null;
            IndexedWord gov2 = null;
            while (matcher.find()) {
                if (w1.equals(matcher.getNode("w1")) && w2.equals(matcher.getNode("w2"))) {
                    gov = matcher.getNode("gov");
                    gov2 = matcher.getNode("gov2");
                    break;
                }
            }
            if (gov2 == null) {
                continue;
            }
            /* Attach the head of the prepositional phrase to
         * the head of w1. */
            if (sg.getRoots().contains(w1)) {
                SemanticGraphEdge edge = sg.getEdge(w1, gov2);
                if (edge == null) {
                    continue;
                }
                sg.removeEdge(edge);
                sg.getRoots().remove(w1);
                sg.addRoot(gov2);
            } else {
                SemanticGraphEdge edge = sg.getEdge(w1, gov2);
                if (edge == null) {
                    continue;
                }
                sg.removeEdge(edge);
                gov = gov == null ? sg.getParent(w1) : gov;
                if (gov == null) {
                    continue;
                }
                /* Determine the relation to use. If it is a relation that can
           * join two clauses and w1 is the head of a copular construction, then
           * use the relation of w1 and its parent. Otherwise use the relation of edge. */
                GrammaticalRelation reln = edge.getRelation();
                if (sg.hasChildWithReln(w1, COPULA)) {
                    GrammaticalRelation reln2 = sg.getEdge(gov, w1).getRelation();
                    if (clauseRelations.contains(reln2)) {
                        reln = reln2;
                    }
                }
                sg.addEdge(gov, gov2, reln, Double.NEGATIVE_INFINITY, false);
            }
            /* Make children of w1 dependents of gov2. */
            for (SemanticGraphEdge edge2 : sg.getOutEdgesSorted(w1)) {
                sg.removeEdge(edge2);
                sg.addEdge(gov2, edge2.getDependent(), edge2.getRelation(), edge2.getWeight(), edge2.isExtra());
            }
            createMultiWordExpression(sg, gov2, CASE_MARKER, w1, w2);
        }
    }
}

Example 9 with IndexedWord

use of edu.stanford.nlp.ling.IndexedWord in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method processNames.

/**
   *
   * Looks for NPs that should have the {@code name} relation and
   * a) changes the structure such that the leftmost token becomes the head
   * b) changes the relation from {@code compound} to {@code name}.
   *
   * Requires NER tags.
   *
   * @param sg A semantic graph.
   */
private static void processNames(SemanticGraph sg) {
    if (!USE_NAME) {
        return;
    }
    // check whether NER tags are available
    IndexedWord rootToken = sg.getFirstRoot();
    if (rootToken == null || !rootToken.containsKey(CoreAnnotations.NamedEntityTagAnnotation.class)) {
        return;
    }
    SemanticGraph sgCopy = sg.makeSoftCopy();
    for (SemgrexPattern pattern : NAME_PATTERNS) {
        SemgrexMatcher matcher = pattern.matcher(sgCopy);
        List<IndexedWord> nameParts = new ArrayList<>();
        IndexedWord head = null;
        while (matcher.find()) {
            IndexedWord w1 = matcher.getNode("w1");
            IndexedWord w2 = matcher.getNode("w2");
            if (head != w1) {
                if (head != null) {
                    processNamesHelper(sg, head, nameParts);
                    nameParts = new ArrayList<>();
                }
                head = w1;
            }
            if (w2.ner().equals(w1.ner())) {
                nameParts.add(w2);
            }
        }
        if (head != null) {
            processNamesHelper(sg, head, nameParts);
            sgCopy = sg.makeSoftCopy();
        }
    }
}

Example 10 with IndexedWord

use of edu.stanford.nlp.ling.IndexedWord in project CoreNLP by stanfordnlp.

the class NaturalLogicWeightsITest method mkSegment.

private Pair<SemanticGraphEdge, List<SemanticGraphEdge>> mkSegment(String root, Pair<String, String>... outEdges) {
    IndexedWord rootVertex = new IndexedWord(mockWord(root));
    List<SemanticGraphEdge> edges = Arrays.asList(outEdges).stream().map(pair -> new SemanticGraphEdge(rootVertex, new IndexedWord(mockWord(pair.second)), GrammaticalRelation.valueOf(Language.English, pair.first), Double.NEGATIVE_INFINITY, false)).collect(Collectors.toList());
    return Pair.makePair(edges.get(0), edges);
}