Examples with IntPair edu.stanford.nlp.util.IntPair used on opensource projects

Example 26 with IntPair

use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.

the class CorefMentionFinder method addNamedEntityStrings.

protected static void addNamedEntityStrings(CoreMap s, Set<String> neStrings, Set<IntPair> namedEntitySpanSet) {
    List<CoreLabel> tokens = s.get(TokensAnnotation.class);
    for (IntPair p : namedEntitySpanSet) {
        StringBuilder sb = new StringBuilder();
        for (int idx = p.get(0); idx < p.get(1); idx++) {
            sb.append(tokens.get(idx).word()).append(" ");
        }
        String str = sb.toString().trim();
        if (str.endsWith(" 's")) {
            str = str.substring(0, str.length() - 3);
        }
        neStrings.add(str);
    }
}

Example 27 with IntPair

use of edu.stanford.nlp.util.IntPair 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 28 with IntPair

use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.

the class DependencyCorefMentionFinder method getNPSpan.

/**
   *  return the left and right most node except copula relation (nsubj & cop) and some others (maybe discourse?)
   *  e.g., you are the person -> return "the person"
   */
private IntPair getNPSpan(IndexedWord headword, SemanticGraph dep, List<CoreLabel> sent) {
    int headwordIdx = headword.index() - 1;
    List<IndexedWord> children = dep.getChildList(headword);
    //    if(children.size()==0) return new IntPair(headwordIdx, headwordIdx);    // the headword is the only word
    // check if we have copula relation
    IndexedWord cop = dep.getChildWithReln(headword, UniversalEnglishGrammaticalRelations.COPULA);
    int startIdx = (cop == null) ? 0 : children.indexOf(cop) + 1;
    // children which will be inside of NP
    List<IndexedWord> insideNP = Generics.newArrayList();
    for (int i = startIdx; i < children.size(); i++) {
        IndexedWord child = children.get(i);
        SemanticGraphEdge edge = dep.getEdge(headword, child);
        if (edge.getRelation().getShortName().matches("dep|discourse|punct")) {
            // skip
            continue;
        } else {
            insideNP.add(child);
        }
    }
    // the headword is the only word
    if (insideNP.size() == 0)
        return new IntPair(headwordIdx, headwordIdx);
    Pair<IndexedWord, IndexedWord> firstChildLeftRight = SemanticGraphUtils.leftRightMostChildVertices(insideNP.get(0), dep);
    Pair<IndexedWord, IndexedWord> lastChildLeftRight = SemanticGraphUtils.leftRightMostChildVertices(insideNP.get(insideNP.size() - 1), dep);
    // headword can be first or last word
    int beginIdx = Math.min(headwordIdx, firstChildLeftRight.first.index() - 1);
    int endIdx = Math.max(headwordIdx, lastChildLeftRight.second.index() - 1);
    return new IntPair(beginIdx, endIdx);
}

Example 29 with IntPair

use of edu.stanford.nlp.util.IntPair 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 30 with IntPair

use of edu.stanford.nlp.util.IntPair 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;
}