Examples with Mention edu.stanford.nlp.coref.data.Mention used on opensource projects

Example 51 with Mention

use of edu.stanford.nlp.coref.data.Mention in project CoreNLP by stanfordnlp.

the class CorefMentionFinder method extractNamedEntityModifiers.

// extract mentions which have same string as another stand-alone mention
protected static void extractNamedEntityModifiers(List<CoreMap> sentences, List<Set<IntPair>> mentionSpanSetList, List<List<Mention>> predictedMentions, Set<String> neStrings) {
    for (int i = 0, sz = sentences.size(); i < sz; i++) {
        List<Mention> mentions = predictedMentions.get(i);
        CoreMap sent = sentences.get(i);
        List<CoreLabel> tokens = sent.get(TokensAnnotation.class);
        Set<IntPair> mentionSpanSet = mentionSpanSetList.get(i);
        for (int j = 0, tSize = tokens.size(); j < tSize; j++) {
            for (String ne : neStrings) {
                int len = ne.split(" ").length;
                if (j + len > tokens.size())
                    continue;
                StringBuilder sb = new StringBuilder();
                for (int k = 0; k < len; k++) {
                    sb.append(tokens.get(k + j).word()).append(" ");
                }
                String phrase = sb.toString().trim();
                int beginIndex = j;
                int endIndex = j + len;
                // include "'s" if it belongs to this named entity
                if (endIndex < tokens.size() && tokens.get(endIndex).word().equals("'s") && tokens.get(endIndex).tag().equals("POS")) {
                    Tree tree = sent.get(TreeAnnotation.class);
                    Tree sToken = tree.getLeaves().get(beginIndex);
                    Tree eToken = tree.getLeaves().get(endIndex);
                    Tree join = tree.joinNode(sToken, eToken);
                    Tree sJoin = join.getLeaves().get(0);
                    Tree eJoin = join.getLeaves().get(join.getLeaves().size() - 1);
                    if (sToken == sJoin && eToken == eJoin) {
                        endIndex++;
                    }
                }
                // include DT if it belongs to this named entity
                if (beginIndex > 0 && tokens.get(beginIndex - 1).tag().equals("DT")) {
                    Tree tree = sent.get(TreeAnnotation.class);
                    Tree sToken = tree.getLeaves().get(beginIndex - 1);
                    Tree eToken = tree.getLeaves().get(endIndex - 1);
                    Tree join = tree.joinNode(sToken, eToken);
                    Tree sJoin = join.getLeaves().get(0);
                    Tree eJoin = join.getLeaves().get(join.getLeaves().size() - 1);
                    if (sToken == sJoin && eToken == eJoin) {
                        beginIndex--;
                    }
                }
                IntPair span = new IntPair(beginIndex, endIndex);
                if (phrase.equalsIgnoreCase(ne) && !mentionSpanSet.contains(span)) {
                    int dummyMentionId = -1;
                    Mention m = new Mention(dummyMentionId, beginIndex, endIndex, tokens, sent.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class), sent.get(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation.class) != null ? sent.get(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation.class) : sent.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class), new ArrayList<>(tokens.subList(beginIndex, endIndex)));
                    mentions.add(m);
                    mentionSpanSet.add(span);
                }
            }
        }
    }
}

Example 52 with Mention

use of edu.stanford.nlp.coref.data.Mention in project CoreNLP by stanfordnlp.

the class CorefMentionFinder method extractPremarkedEntityMentions.

protected static void extractPremarkedEntityMentions(CoreMap s, List<Mention> mentions, Set<IntPair> mentionSpanSet, Set<IntPair> namedEntitySpanSet) {
    List<CoreLabel> sent = s.get(CoreAnnotations.TokensAnnotation.class);
    SemanticGraph basicDependency = s.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
    SemanticGraph enhancedDependency = s.get(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation.class);
    if (enhancedDependency == null) {
        enhancedDependency = s.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
    }
    int beginIndex = -1;
    for (CoreLabel w : sent) {
        MultiTokenTag t = w.get(CoreAnnotations.MentionTokenAnnotation.class);
        if (t != null) {
            // Part of a mention
            if (t.isStart()) {
                // Start of mention
                beginIndex = w.get(CoreAnnotations.IndexAnnotation.class) - 1;
            }
            if (t.isEnd()) {
                // end of mention
                int endIndex = w.get(CoreAnnotations.IndexAnnotation.class);
                if (beginIndex >= 0) {
                    IntPair mSpan = new IntPair(beginIndex, endIndex);
                    int dummyMentionId = -1;
                    Mention m = new Mention(dummyMentionId, beginIndex, endIndex, sent, basicDependency, enhancedDependency, new ArrayList<>(sent.subList(beginIndex, endIndex)));
                    mentions.add(m);
                    mentionSpanSet.add(mSpan);
                    beginIndex = -1;
                } else {
                    Redwood.log("Start of marked mention not found in sentence: " + t + " at tokenIndex=" + (w.get(CoreAnnotations.IndexAnnotation.class) - 1) + " for " + s.get(CoreAnnotations.TextAnnotation.class));
                }
            }
        }
    }
}

Example 53 with Mention

use of edu.stanford.nlp.coref.data.Mention in project CoreNLP by stanfordnlp.

the class DependencyCorefMentionFinder method extractPronounForHeadword.

private void extractPronounForHeadword(IndexedWord headword, SemanticGraph dep, CoreMap s, List<Mention> mentions, Set<IntPair> mentionSpanSet, Set<IntPair> namedEntitySpanSet) {
    List<CoreLabel> sent = s.get(CoreAnnotations.TokensAnnotation.class);
    SemanticGraph basic = s.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
    SemanticGraph enhanced = s.get(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation.class);
    if (enhanced == null) {
        enhanced = s.get(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class);
    }
    int beginIdx = headword.index() - 1;
    int endIdx = headword.index();
    // handle "you all", "they both" etc
    if (sent.size() > headword.index() && sent.get(headword.index()).word().matches("all|both")) {
        IndexedWord c = dep.getNodeByIndex(headword.index() + 1);
        SemanticGraphEdge edge = dep.getEdge(headword, c);
        if (edge != null)
            endIdx++;
    }
    IntPair mSpan = new IntPair(beginIdx, endIdx);
    if (!mentionSpanSet.contains(mSpan) && (!insideNE(mSpan, namedEntitySpanSet))) {
        int dummyMentionId = -1;
        Mention m = new Mention(dummyMentionId, beginIdx, endIdx, sent, basic, enhanced, new ArrayList<>(sent.subList(beginIdx, endIdx)));
        m.headIndex = headword.index() - 1;
        m.headWord = sent.get(m.headIndex);
        m.headString = m.headWord.word().toLowerCase(Locale.ENGLISH);
        mentions.add(m);
        mentionSpanSet.add(mSpan);
    }
    // when pronoun is a part of conjunction (e.g., you and I)
    Set<IndexedWord> conjChildren = dep.getChildrenWithReln(headword, UniversalEnglishGrammaticalRelations.CONJUNCT);
    if (conjChildren.size() > 0) {
        IntPair npSpan = getNPSpan(headword, dep, sent);
        beginIdx = npSpan.get(0);
        endIdx = npSpan.get(1) + 1;
        // try not to have span that ends with ,
        if (",".equals(sent.get(endIdx - 1).word())) {
            endIdx--;
        }
        addMention(beginIdx, endIdx, headword, mentions, mentionSpanSet, namedEntitySpanSet, sent, basic, enhanced);
    }
}

Example 54 with Mention

use of edu.stanford.nlp.coref.data.Mention in project CoreNLP by stanfordnlp.

the class DependencyCorefMentionFinder method findMentions.

/** Main method of mention detection.
   *  Extract all NP, PRP or NE, and filter out by manually written patterns.
   */
@Override
public List<List<Mention>> findMentions(Annotation doc, Dictionaries dict, Properties props) {
    List<List<Mention>> predictedMentions = new ArrayList<>();
    Set<String> neStrings = Generics.newHashSet();
    List<Set<IntPair>> mentionSpanSetList = Generics.newArrayList();
    List<CoreMap> sentences = doc.get(CoreAnnotations.SentencesAnnotation.class);
    for (CoreMap s : sentences) {
        List<Mention> mentions = new ArrayList<>();
        predictedMentions.add(mentions);
        Set<IntPair> mentionSpanSet = Generics.newHashSet();
        Set<IntPair> namedEntitySpanSet = Generics.newHashSet();
        extractPremarkedEntityMentions(s, mentions, mentionSpanSet, namedEntitySpanSet);
        HybridCorefMentionFinder.extractNamedEntityMentions(s, mentions, mentionSpanSet, namedEntitySpanSet);
        extractNPorPRPFromDependency(s, mentions, mentionSpanSet, namedEntitySpanSet);
        addNamedEntityStrings(s, neStrings, namedEntitySpanSet);
        mentionSpanSetList.add(mentionSpanSet);
    }
    for (int i = 0; i < sentences.size(); i++) {
        findHead(sentences.get(i), predictedMentions.get(i));
    }
    // mention selection based on document-wise info
    removeSpuriousMentions(doc, predictedMentions, dict, CorefProperties.removeNestedMentions(props), lang);
    // if this is for MD training, skip classification
    if (!CorefProperties.isMentionDetectionTraining(props)) {
        mdClassifier.classifyMentions(predictedMentions, dict, props);
    }
    return predictedMentions;
}

Example 55 with Mention

use of edu.stanford.nlp.coref.data.Mention in project CoreNLP by stanfordnlp.

the class DependencyCorefMentionFinder method addMention.

private void addMention(int beginIdx, int endIdx, IndexedWord headword, List<Mention> mentions, Set<IntPair> mentionSpanSet, Set<IntPair> namedEntitySpanSet, List<CoreLabel> sent, SemanticGraph basic, SemanticGraph enhanced) {
    IntPair mSpan = new IntPair(beginIdx, endIdx);
    if (!mentionSpanSet.contains(mSpan) && (!insideNE(mSpan, namedEntitySpanSet))) {
        int dummyMentionId = -1;
        Mention m = new Mention(dummyMentionId, beginIdx, endIdx, sent, basic, enhanced, new ArrayList<>(sent.subList(beginIdx, endIdx)));
        m.headIndex = headword.index() - 1;
        m.headWord = sent.get(m.headIndex);
        m.headString = m.headWord.word().toLowerCase(Locale.ENGLISH);
        mentions.add(m);
        mentionSpanSet.add(mSpan);
    }
}