Examples with SemanticGraph edu.stanford.nlp.semgraph.SemanticGraph used on opensource projects

Example 96 with SemanticGraph

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

the class SsurgeonPattern method execute.

/**
   * Executes the Ssurgeon edit, but with the given Semgrex Pattern, instead of the one attached to this
   * pattern.
   * 
   * NOTE: Predicate tests are still active here, and any named nodes required for evaluation must be
   * present.
   */
public Collection<SemanticGraph> execute(SemanticGraph sg, SemgrexPattern overridePattern) throws Exception {
    SemgrexMatcher matcher = overridePattern.matcher(sg);
    Collection<SemanticGraph> generated = new ArrayList<>();
    while (matcher.find()) {
        if (predicateTest != null) {
            if (!predicateTest.test(matcher))
                continue;
        }
        // We reset the named node map with each edit set, since these edits
        // should exist in a separate graph for each unique Semgrex match.
        nodeMap = Generics.newHashMap();
        SemanticGraph tgt = new SemanticGraph(sg);
        for (SsurgeonEdit edit : editScript) {
            edit.evaluate(tgt, matcher);
        }
        generated.add(tgt);
    }
    return generated;
}

Example 97 with SemanticGraph

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

the class EnglishGrammaticalStructure 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 th 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 98 with SemanticGraph

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

the class UniversalEnglishGrammaticalStructure method treatCC.

private static void treatCC(SemanticGraph sg) {
    // Construct a map from tree nodes to the set of typed
    // dependencies in which the node appears as dependent.
    Map<IndexedWord, Set<SemanticGraphEdge>> map = Generics.newHashMap();
    // Construct a map of tree nodes being governor of a subject grammatical
    // relation to that relation
    Map<IndexedWord, SemanticGraphEdge> subjectMap = Generics.newHashMap();
    // Construct a set of TreeGraphNodes with a passive auxiliary on them
    Set<IndexedWord> withPassiveAuxiliary = Generics.newHashSet();
    // Construct a map of tree nodes being governor of an object grammatical
    // relation to that relation
    // Map<TreeGraphNode, TypedDependency> objectMap = new
    // HashMap<TreeGraphNode, TypedDependency>();
    List<IndexedWord> rcmodHeads = Generics.newArrayList();
    List<IndexedWord> prepcDep = Generics.newArrayList();
    for (SemanticGraphEdge edge : sg.edgeIterable()) {
        if (!map.containsKey(edge.getDependent())) {
            // NB: Here and in other places below, we use a TreeSet (which extends
            // SortedSet) to guarantee that results are deterministic)
            map.put(edge.getDependent(), new TreeSet<>());
        }
        map.get(edge.getDependent()).add(edge);
        if (edge.getRelation().equals(AUX_PASSIVE_MODIFIER)) {
            withPassiveAuxiliary.add(edge.getGovernor());
        }
        // look for subjects
        if (edge.getRelation().getParent() == NOMINAL_SUBJECT || edge.getRelation().getParent() == SUBJECT || edge.getRelation().getParent() == CLAUSAL_SUBJECT) {
            if (!subjectMap.containsKey(edge.getGovernor())) {
                subjectMap.put(edge.getGovernor(), edge);
            }
        }
        // look for rcmod relations
        if (edge.getRelation() == RELATIVE_CLAUSE_MODIFIER) {
            rcmodHeads.add(edge.getGovernor());
        }
        // to avoid wrong propagation of dobj
        if (edge.getRelation().toString().startsWith("acl:") || edge.getRelation().toString().startsWith("advcl:")) {
            prepcDep.add(edge.getDependent());
        }
    }
    // log.info(map);
    // if (DEBUG) log.info("Subject map: " + subjectMap);
    // if (DEBUG) log.info("Object map: " + objectMap);
    // log.info(rcmodHeads);
    // create a new list of typed dependencies
    //Collection<TypedDependency> newTypedDeps = new ArrayList<TypedDependency>(list);
    SemanticGraph sgCopy = sg.makeSoftCopy();
    // find typed deps of form conj(gov,dep)
    for (SemanticGraphEdge edge : sgCopy.edgeIterable()) {
        if (UniversalEnglishGrammaticalRelations.getConjs().contains(edge.getRelation())) {
            IndexedWord gov = edge.getGovernor();
            IndexedWord dep = edge.getDependent();
            // look at the dep in the conjunct
            Set<SemanticGraphEdge> gov_relations = map.get(gov);
            // log.info("gov " + gov);
            if (gov_relations != null) {
                for (SemanticGraphEdge edge1 : gov_relations) {
                    // log.info("gov rel " + td1);
                    IndexedWord newGov = edge1.getGovernor();
                    // is possible to have overlapping newGov & dep
                    if (newGov.equals(dep)) {
                        continue;
                    }
                    GrammaticalRelation newRel = edge1.getRelation();
                    //TODO: Do we want to copy case markers here?
                    if (newRel != ROOT && newRel != CASE_MARKER) {
                        if (rcmodHeads.contains(gov) && rcmodHeads.contains(dep)) {
                            // to prevent wrong propagation in the case of long dependencies in relative clauses
                            if (newRel != DIRECT_OBJECT && newRel != NOMINAL_SUBJECT) {
                                if (DEBUG) {
                                    log.info("Adding new " + newRel + " dependency from " + newGov + " to " + dep + " (subj/obj case)");
                                }
                                sg.addEdge(newGov, dep, newRel, Double.NEGATIVE_INFINITY, true);
                            }
                        } else {
                            if (DEBUG) {
                                log.info("Adding new " + newRel + " dependency from " + newGov + " to " + dep);
                            }
                            sg.addEdge(newGov, dep, newRel, Double.NEGATIVE_INFINITY, true);
                        }
                    }
                }
            }
            // propagate subjects
            // look at the gov in the conjunct: if it is has a subject relation,
            // the dep is a verb and the dep doesn't have a subject relation
            // then we want to add a subject relation for the dep.
            // (By testing for the dep to be a verb, we are going to miss subject of
            // copular verbs! but
            // is it safe to relax this assumption?? i.e., just test for the subject
            // part)
            // CDM 2008: I also added in JJ, since participial verbs are often
            // tagged JJ
            String tag = dep.tag();
            if (subjectMap.containsKey(gov) && (tag.startsWith("VB") || tag.startsWith("JJ")) && !subjectMap.containsKey(dep)) {
                SemanticGraphEdge tdsubj = subjectMap.get(gov);
                // check for wrong nsubjpass: if the new verb is VB or VBZ or VBP or JJ, then
                // add nsubj (if it is tagged correctly, should do this for VBD too, but we don't)
                GrammaticalRelation relation = tdsubj.getRelation();
                if (relation == NOMINAL_PASSIVE_SUBJECT) {
                    if (isDefinitelyActive(tag)) {
                        relation = NOMINAL_SUBJECT;
                    }
                } else if (relation == CLAUSAL_PASSIVE_SUBJECT) {
                    if (isDefinitelyActive(tag)) {
                        relation = CLAUSAL_SUBJECT;
                    }
                } else if (relation == NOMINAL_SUBJECT) {
                    if (withPassiveAuxiliary.contains(dep)) {
                        relation = NOMINAL_PASSIVE_SUBJECT;
                    }
                } else if (relation == CLAUSAL_SUBJECT) {
                    if (withPassiveAuxiliary.contains(dep)) {
                        relation = CLAUSAL_PASSIVE_SUBJECT;
                    }
                }
                if (DEBUG) {
                    log.info("Adding new " + relation + " dependency from " + dep + " to " + tdsubj.getDependent() + " (subj propagation case)");
                }
                sg.addEdge(dep, tdsubj.getDependent(), relation, Double.NEGATIVE_INFINITY, true);
            }
        // propagate objects
        // cdm july 2010: This bit of code would copy a dobj from the first
        // clause to a later conjoined clause if it didn't
        // contain its own dobj or prepc. But this is too aggressive and wrong
        // if the later clause is intransitive
        // (including passivized cases) and so I think we have to not have this
        // done always, and see no good "sometimes" heuristic.
        // IF WE WERE TO REINSTATE, SHOULD ALSO NOT ADD OBJ IF THERE IS A ccomp
        // (SBAR).
        // if (objectMap.containsKey(gov) &&
        // dep.tag().startsWith("VB") && ! objectMap.containsKey(dep)
        // && ! prepcDep.contains(gov)) {
        // TypedDependency tdobj = objectMap.get(gov);
        // if (DEBUG) {
        // log.info("Adding new " + tdobj.reln() + " dependency from "
        // + dep + " to " + tdobj.dep() + " (obj propagation case)");
        // }
        // newTypedDeps.add(new TypedDependency(tdobj.reln(), dep,
        // tdobj.dep()));
        // }
        }
    }
}

Example 99 with SemanticGraph

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

the class UniversalEnglishGrammaticalStructure method correctSubjPass.

/**
   * This method corrects subjects of verbs for which we identified an auxpass,
   * but didn't identify the subject as passive.
   *
   * @param sg SemanticGraph to work on
   */
private static void correctSubjPass(SemanticGraph sg) {
    /* If the graph doesn't have a root (most likely because
     * a parsing error, we can't match Semgrexes, so do
     * nothing. */
    if (sg.getRoots().isEmpty())
        return;
    SemanticGraph sgCopy = sg.makeSoftCopy();
    SemgrexMatcher matcher = CORRECT_SUBJPASS_PATTERN.matcher(sgCopy);
    while (matcher.find()) {
        IndexedWord gov = matcher.getNode("gov");
        IndexedWord subj = matcher.getNode("subj");
        SemanticGraphEdge edge = sg.getEdge(gov, subj);
        GrammaticalRelation reln = null;
        if (edge.getRelation() == NOMINAL_SUBJECT) {
            reln = NOMINAL_PASSIVE_SUBJECT;
        } else if (edge.getRelation() == CLAUSAL_SUBJECT) {
            reln = CLAUSAL_PASSIVE_SUBJECT;
        } else if (edge.getRelation() == CONTROLLING_NOMINAL_SUBJECT) {
            reln = CONTROLLING_NOMINAL_PASSIVE_SUBJECT;
        } else if (edge.getRelation() == CONTROLLING_CLAUSAL_SUBJECT) {
            reln = CONTROLLING_CLAUSAL_PASSIVE_SUBJECT;
        }
        if (reln != null) {
            sg.removeEdge(edge);
            sg.addEdge(gov, subj, reln, Double.NEGATIVE_INFINITY, false);
        }
    }
}

Example 100 with SemanticGraph

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

the class UniversalEnglishGrammaticalStructure method demoteQuantificationalModifiers.

private static void demoteQuantificationalModifiers(SemanticGraph sg) {
    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);
        }
    }
}