Examples with SemgrexPattern edu.stanford.nlp.semgraph.semgrex.SemgrexPattern used on opensource projects

Example 11 with SemgrexPattern

use of edu.stanford.nlp.semgraph.semgrex.SemgrexPattern in project CoreNLP by stanfordnlp.

the class RelationTripleSegmenterTest method testVPOnlyReplacedWith.

public void testVPOnlyReplacedWith() {
    String conll = "1\treplaced\t0\tconj:and\tVBD\n" + "2\twith\t5\tcase\tIN\n" + "3\ta\t5\tdet\tDT\n" + "4\tdifferent\t5\tamod\tJJ\n" + "5\ttype\t1\tnmod:with\tNN\n" + "6\tof\t7\tcase\tIN\n" + "7\tfilter\t5\tnmod:of\tNN\n";
    // Positive case
    boolean matches = false;
    SemanticGraph tree = mkTree(conll).first;
    for (SemgrexPattern candidate : new RelationTripleSegmenter().VP_PATTERNS) {
        if (candidate.matcher(tree).matches()) {
            matches = true;
        }
    }
    assertTrue(matches);
}

Example 12 with SemgrexPattern

use of edu.stanford.nlp.semgraph.semgrex.SemgrexPattern in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method demoteQuantificationalModifiers.

private static void demoteQuantificationalModifiers(SemanticGraph sg) {
    /* Semgrexes require a graph with a root. */
    if (sg.getRoots().isEmpty()) {
        return;
    }
    SemanticGraph sgCopy = sg.makeSoftCopy();
    SemgrexMatcher matcher = QUANT_MOD_3W_PATTERN.matcher(sgCopy);
    while (matcher.findNextMatchingNode()) {
        IndexedWord w1 = matcher.getNode("w1");
        IndexedWord w2 = matcher.getNode("w2");
        IndexedWord w3 = matcher.getNode("w3");
        IndexedWord gov = matcher.getNode("gov");
        demoteQmodParentHelper(sg, gov, w2);
        List<IndexedWord> otherDeps = Generics.newLinkedList();
        otherDeps.add(w1);
        otherDeps.add(w2);
        otherDeps.add(w3);
        demoteQmodMWEHelper(sg, otherDeps, gov, w2);
    }
    for (SemgrexPattern p : QUANT_MOD_2W_PATTERNS) {
        sgCopy = sg.makeSoftCopy();
        matcher = p.matcher(sgCopy);
        while (matcher.findNextMatchingNode()) {
            IndexedWord w1 = matcher.getNode("w1");
            IndexedWord w2 = matcher.getNode("w2");
            IndexedWord gov = matcher.getNode("gov");
            demoteQmodParentHelper(sg, gov, w1);
            List<IndexedWord> otherDeps = Generics.newLinkedList();
            otherDeps.add(w1);
            otherDeps.add(w2);
            demoteQmodMWEHelper(sg, otherDeps, gov, w1);
        }
    }
}

Example 13 with SemgrexPattern

use of edu.stanford.nlp.semgraph.semgrex.SemgrexPattern in project CoreNLP by stanfordnlp.

the class UniversalEnglishGrammaticalStructure method addCaseMarkerInformation.

/**
 * Adds the case marker(s) to all nmod, acl and advcl relations that are
 * modified by one or more case markers(s).
 *
 * @param enhanceOnlyNmods If this is set to true, then prepositions will only be appended to nmod
 *                         relations (and not to acl or advcl) relations.
 *
 * @see UniversalEnglishGrammaticalStructure#addCaseMarkersToReln
 */
private static void addCaseMarkerInformation(SemanticGraph sg, boolean enhanceOnlyNmods) {
    /* Semgrexes require a graph with a root. */
    if (sg.getRoots().isEmpty())
        return;
    /* passive agent */
    SemanticGraph sgCopy = sg.makeSoftCopy();
    SemgrexMatcher matcher = PASSIVE_AGENT_PATTERN.matcher(sgCopy);
    while (matcher.find()) {
        IndexedWord caseMarker = matcher.getNode("c1");
        IndexedWord gov = matcher.getNode("gov");
        IndexedWord mod = matcher.getNode("mod");
        addPassiveAgentToReln(sg, gov, mod, caseMarker);
    }
    List<IndexedWord> oldCaseMarkers = Generics.newArrayList();
    /* 3-word prepositions */
    for (SemgrexPattern p : PREP_MW3_PATTERNS) {
        sgCopy = sg.makeSoftCopy();
        matcher = p.matcher(sgCopy);
        while (matcher.find()) {
            if (enhanceOnlyNmods && !matcher.getRelnString("reln").equals("nmod") && !matcher.getRelnString("reln").equals("obl")) {
                continue;
            }
            List<IndexedWord> caseMarkers = Generics.newArrayList(3);
            caseMarkers.add(matcher.getNode("c1"));
            caseMarkers.add(matcher.getNode("c2"));
            caseMarkers.add(matcher.getNode("c3"));
            Collections.sort(caseMarkers);
            /* We only want to match every case marker once. */
            if (caseMarkers.equals(oldCaseMarkers))
                continue;
            IndexedWord gov = matcher.getNode("gov");
            IndexedWord mod = matcher.getNode("mod");
            addCaseMarkersToReln(sg, gov, mod, caseMarkers);
            oldCaseMarkers = caseMarkers;
        }
    }
    /* 2-word prepositions */
    for (SemgrexPattern p : PREP_MW2_PATTERNS) {
        sgCopy = sg.makeSoftCopy();
        matcher = p.matcher(sgCopy);
        while (matcher.find()) {
            if (enhanceOnlyNmods && !matcher.getRelnString("reln").equals("nmod") && !matcher.getRelnString("reln").equals("obl")) {
                continue;
            }
            List<IndexedWord> caseMarkers = Generics.newArrayList(2);
            caseMarkers.add(matcher.getNode("c1"));
            caseMarkers.add(matcher.getNode("c2"));
            Collections.sort(caseMarkers);
            /* We only want to match every case marker once. */
            if (caseMarkers.equals(oldCaseMarkers))
                continue;
            IndexedWord gov = matcher.getNode("gov");
            IndexedWord mod = matcher.getNode("mod");
            addCaseMarkersToReln(sg, gov, mod, caseMarkers);
            oldCaseMarkers = caseMarkers;
        }
    }
    /* Single-word prepositions */
    for (SemgrexPattern p : PREP_PATTERNS) {
        sgCopy = sg.makeSoftCopy();
        matcher = p.matcher(sgCopy);
        while (matcher.find()) {
            if (enhanceOnlyNmods && !matcher.getRelnString("reln").equals("nmod") && !matcher.getRelnString("reln").equals("obl")) {
                continue;
            }
            List<IndexedWord> caseMarkers = Generics.newArrayList(1);
            caseMarkers.add(matcher.getNode("c1"));
            if (caseMarkers.equals(oldCaseMarkers))
                continue;
            IndexedWord gov = matcher.getNode("gov");
            IndexedWord mod = matcher.getNode("mod");
            addCaseMarkersToReln(sg, gov, mod, caseMarkers);
            oldCaseMarkers = caseMarkers;
        }
    }
}

Example 14 with SemgrexPattern

use of edu.stanford.nlp.semgraph.semgrex.SemgrexPattern in project CoreNLP by stanfordnlp.

the class KBPSemgrexExtractor method matches.

/**
 * Returns whether any of the given patterns match this tree.
 */
private boolean matches(CoreMap sentence, Collection<SemgrexPattern> rulesForRel, KBPInput input, SemanticGraph graph) {
    if (graph == null || graph.isEmpty()) {
        return false;
    }
    List<CoreLabel> tokens = sentence.get(CoreAnnotations.TokensAnnotation.class);
    for (int i : input.subjectSpan) {
        if ("O".equals(tokens.get(i).ner())) {
            tokens.get(i).setNER(input.subjectType.name);
        }
    }
    for (int i : input.objectSpan) {
        if ("O".equals(tokens.get(i).ner())) {
            tokens.get(i).setNER(input.objectType.name);
        }
    }
    for (SemgrexPattern p : rulesForRel) {
        SemgrexMatcher n = p.matcher(graph);
        while (n.find()) {
            IndexedWord entity = n.getNode("entity");
            IndexedWord slot = n.getNode("slot");
            if (entity == null) {
                // really this is a hideous bug, right?  these rules
                // should all have entity and slot set
                logger.warn("Found a relation with a missing entity: " + p);
                continue;
            }
            if (slot == null) {
                logger.warn("Found a relation with a missing slot: " + p);
                continue;
            }
            boolean hasSubject = entity.index() >= input.subjectSpan.start() + 1 && entity.index() <= input.subjectSpan.end();
            boolean hasObject = slot.index() >= input.objectSpan.start() + 1 && slot.index() <= input.objectSpan.end();
            if (hasSubject && hasObject) {
                return true;
            }
        }
    }
    return false;
}

Example 15 with SemgrexPattern

use of edu.stanford.nlp.semgraph.semgrex.SemgrexPattern in project CoreNLP by stanfordnlp.

the class KBPSemgrexExtractor method classify.

@Override
public Pair<String, Double> classify(KBPInput input) {
    for (RelationType rel : RelationType.values()) {
        if (rules.containsKey(rel) && rel.entityType == input.subjectType && rel.validNamedEntityLabels.contains(input.objectType)) {
            Collection<SemgrexPattern> rulesForRel = rules.get(rel);
            CoreMap sentence = input.sentence.asCoreMap(Sentence::nerTags, Sentence::dependencyGraph);
            boolean matches = (matches(sentence, rulesForRel, input, sentence.get(SemanticGraphCoreAnnotations.EnhancedPlusPlusDependenciesAnnotation.class)) || matches(sentence, rulesForRel, input, sentence.get(SemanticGraphCoreAnnotations.AlternativeDependenciesAnnotation.class)));
            if (matches) {
                // logger.log("MATCH for " + rel +  ".  sentence:" + sentence + " with rules for  " + rel);
                return Pair.makePair(rel.canonicalName, 1.0);
            }
        }
    }
    return Pair.makePair(NO_RELATION, 1.0);
}