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

Example 46 with Mention

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

the class ProtobufAnnotationSerializer method fromProtoNoTokens.

private Mention fromProtoNoTokens(CoreNLPProtos.Mention protoMention) {
    Mention returnMention = new Mention();
    // set enums
    if (protoMention.getMentionType() != null && !protoMention.getMentionType().equals("")) {
        returnMention.mentionType = Dictionaries.MentionType.valueOf(protoMention.getMentionType());
    }
    if (protoMention.getNumber() != null && !protoMention.getNumber().equals("")) {
        returnMention.number = Dictionaries.Number.valueOf(protoMention.getNumber());
    }
    if (protoMention.getGender() != null && !protoMention.getGender().equals("")) {
        returnMention.gender = Dictionaries.Gender.valueOf(protoMention.getGender());
    }
    if (protoMention.getAnimacy() != null && !protoMention.getAnimacy().equals("")) {
        returnMention.animacy = Dictionaries.Animacy.valueOf(protoMention.getAnimacy());
    }
    if (protoMention.getPerson() != null && !protoMention.getPerson().equals("")) {
        returnMention.person = Dictionaries.Person.valueOf(protoMention.getPerson());
    }
    // TO DO: if the original Mention had "" for this field it will be lost, should deal with this problem
    if (!protoMention.getHeadString().equals("")) {
        returnMention.headString = protoMention.getHeadString();
    }
    // TO DO: if the original Mention had "" for this field it will be lost, should deal with this problem
    if (!protoMention.getNerString().equals("")) {
        returnMention.nerString = protoMention.getNerString();
    }
    returnMention.startIndex = protoMention.getStartIndex();
    returnMention.endIndex = protoMention.getEndIndex();
    returnMention.headIndex = protoMention.getHeadIndex();
    returnMention.mentionID = protoMention.getMentionID();
    returnMention.originalRef = protoMention.getOriginalRef();
    returnMention.goldCorefClusterID = protoMention.getGoldCorefClusterID();
    returnMention.corefClusterID = protoMention.getCorefClusterID();
    returnMention.mentionNum = protoMention.getMentionNum();
    returnMention.sentNum = protoMention.getSentNum();
    returnMention.utter = protoMention.getUtter();
    returnMention.paragraph = protoMention.getParagraph();
    returnMention.isSubject = protoMention.getIsSubject();
    returnMention.isDirectObject = protoMention.getIsDirectObject();
    returnMention.isIndirectObject = protoMention.getIsIndirectObject();
    returnMention.isPrepositionObject = protoMention.getIsPrepositionObject();
    returnMention.hasTwin = protoMention.getHasTwin();
    returnMention.generic = protoMention.getGeneric();
    returnMention.isSingleton = protoMention.getIsSingleton();
    // handle the sets of Strings
    if (protoMention.getDependentsCount() != 0) {
        returnMention.dependents = new HashSet<>();
        returnMention.dependents.addAll(protoMention.getDependentsList());
    }
    if (protoMention.getPreprocessedTermsCount() != 0) {
        returnMention.preprocessedTerms = new ArrayList<>();
        returnMention.preprocessedTerms.addAll(protoMention.getPreprocessedTermsList());
    }
    return returnMention;
}

Example 47 with Mention

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

the class NameMatch method checkEntityMatch.

public boolean checkEntityMatch(Document document, CorefCluster mentionCluster, CorefCluster potentialAntecedent, Dictionaries dict, Set<Mention> roleSet) {
    Boolean matched = false;
    Mention mainMention = mentionCluster.getRepresentativeMention();
    Mention antMention = potentialAntecedent.getRepresentativeMention();
    // Check if the representative mentions are compatible
    if (isNamedMention(mainMention, dict, roleSet) && isNamedMention(antMention, dict, roleSet)) {
        if (mainMention.originalSpan.size() > minTokens || antMention.originalSpan.size() > minTokens) {
            if (CorefRules.entityAttributesAgree(mentionCluster, potentialAntecedent, ignoreGender)) {
                if (supportedNerTypes.contains(mainMention.nerString) || supportedNerTypes.contains(antMention.nerString)) {
                    matched = mentionMatcher.isCompatible(mainMention, antMention);
                    if (matched != null) {
                        //Redwood.log("Match '" + mainMention + "' with '" + antMention + "' => " + matched);
                        if (!matched) {
                            document.addIncompatible(mainMention, antMention);
                        }
                    } else {
                        matched = false;
                    }
                }
            }
        }
    }
    return matched;
}

Example 48 with Mention

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

the class Sieve method resolveMention.

public String resolveMention(Document document, Dictionaries dict, Properties props) throws Exception {
    StringBuilder sbLog = new StringBuilder();
    if (HybridCorefProperties.debug(props)) {
        sbLog.append("=======================================================");
        sbLog.append(HybridCorefPrinter.printRawDoc(document, true, true));
    }
    for (List<Mention> mentionsInSent : document.predictedMentions) {
        for (int mIdx = 0; mIdx < mentionsInSent.size(); mIdx++) {
            Mention m = mentionsInSent.get(mIdx);
            if (skipMentionType(m, props))
                continue;
            findCoreferentAntecedent(m, mIdx, document, dict, props, sbLog);
        }
    }
    return sbLog.toString();
}

Example 49 with Mention

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

the class CorefMentionFinder method removeSpuriousMentionsZh.

protected void removeSpuriousMentionsZh(Annotation doc, List<List<Mention>> predictedMentions, Dictionaries dict, boolean removeNested) {
    List<CoreMap> sentences = doc.get(CoreAnnotations.SentencesAnnotation.class);
    // this goes through each sentence -- predictedMentions has a list for each sentence
    for (int i = 0, sz = predictedMentions.size(); i < sz; i++) {
        List<Mention> mentions = predictedMentions.get(i);
        List<CoreLabel> sent = sentences.get(i).get(CoreAnnotations.TokensAnnotation.class);
        Set<Mention> remove = Generics.newHashSet();
        for (Mention m : mentions) {
            if (m.headWord.ner().matches("PERCENT|MONEY|QUANTITY|CARDINAL")) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING number NER: " + m.spanToString());
            } else if (m.originalSpan.size() == 1 && m.headWord.tag().equals("CD")) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING number: " + m.spanToString());
            } else if (dict.removeWords.contains(m.spanToString())) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING removeWord: " + m.spanToString());
            } else if (mentionContainsRemoveChars(m, dict.removeChars)) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING removeChars: " + m.spanToString());
            } else if (m.headWord.tag().equals("PU")) {
                // punctuation-only mentions
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Punctuation only mention: " + m.spanToString());
            } else if (mentionIsDemonym(m, dict.countries)) {
                // demonyms -- this seems to be a no-op on devset. Maybe not working?
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Removed demonym: " + m.spanToString());
            } else if (m.spanToString().equals("问题") && m.startIndex > 0 && sent.get(m.startIndex - 1).word().endsWith("没")) {
                // 没 问题 - this is maybe okay but having 问题 on removeWords was dangerous
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Removed meiyou: " + m.spanToString());
            } else if (mentionIsRangren(m, sent)) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Removed rangren: " + m.spanToString());
            } else if (m.spanToString().equals("你") && m.startIndex < sent.size() - 1 && sent.get(m.startIndex + 1).word().startsWith("知道")) {
                // 你 知道
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Removed nizhidao: " + m.spanToString());
            // The words that used to be in this case are now handled more generallyin removeCharsZh
            // } else if (m.spanToString().contains("什么") || m.spanToString().contains("多少")) {
            //   remove.add(m);
            //   if (VERBOSE) log.info("MENTION FILTERING Removed many/few mention ending: " + m.spanToString());
            } else if (m.spanToString().endsWith("的")) {
                remove.add(m);
                if (VERBOSE)
                    log.info("MENTION FILTERING Removed de ending mention: " + m.spanToString());
            // omit this case, it decreases performance. A few useful interrogative pronouns are now in the removeChars list
            // } else if (mentionIsInterrogativePronoun(m, dict.interrogativePronouns)) {
            //     remove.add(m);
            //     if (VERBOSE) log.info("MENTION FILTERING Removed interrogative pronoun: " + m.spanToString());
            }
        // 的 handling
        //        if(m.startIndex>0 && sent.get(m.startIndex-1).word().equals("的")) {
        //          // remove.add(m);
        //          Tree t = sentences.get(i).get(TreeAnnotation.class);
        //          Tree mTree = m.mentionSubTree;
        //          if(mTree==null) continue;
        //          for(Tree p : t.pathNodeToNode(mTree, t)) {
        //            if(mTree==p) continue;
        //            if(p.value().equals("NP")) {
        //              remove.add(m);
        //            }
        //          }
        //        }
        }
        // nested mention with shared headword (except apposition, enumeration): pick larger one
        if (removeNested) {
            for (Mention m1 : mentions) {
                for (Mention m2 : mentions) {
                    if (m1 == m2 || remove.contains(m1) || remove.contains(m2))
                        continue;
                    if (m1.sentNum == m2.sentNum && m1.headWord == m2.headWord && m2.insideIn(m1)) {
                        if (m2.endIndex < sent.size() && (sent.get(m2.endIndex).get(CoreAnnotations.PartOfSpeechAnnotation.class).equals(",") || sent.get(m2.endIndex).get(CoreAnnotations.PartOfSpeechAnnotation.class).equals("CC"))) {
                            continue;
                        }
                        remove.add(m2);
                    }
                }
            }
        }
        mentions.removeAll(remove);
    }
// for each sentence
}

Example 50 with Mention

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

the class CorefMentionFinder method findHead.

public void findHead(CoreMap s, List<Mention> mentions) {
    Tree tree = s.get(TreeCoreAnnotations.TreeAnnotation.class);
    List<CoreLabel> sent = s.get(CoreAnnotations.TokensAnnotation.class);
    tree.indexSpans(0);
    for (Mention m : mentions) {
        if (lang == Locale.CHINESE) {
            findHeadChinese(sent, m);
        } else {
            CoreLabel head = (CoreLabel) findSyntacticHead(m, tree, sent).label();
            m.headIndex = head.get(CoreAnnotations.IndexAnnotation.class) - 1;
            m.headWord = sent.get(m.headIndex);
            m.headString = m.headWord.get(CoreAnnotations.TextAnnotation.class).toLowerCase(Locale.ENGLISH);
        }
        int start = m.headIndex - m.startIndex;
        if (start < 0 || start >= m.originalSpan.size()) {
            Redwood.log("Invalid index for head " + start + "=" + m.headIndex + "-" + m.startIndex + ": originalSpan=[" + StringUtils.joinWords(m.originalSpan, " ") + "], head=" + m.headWord);
            Redwood.log("Setting head string to entire mention");
            m.headIndex = m.startIndex;
            m.headWord = m.originalSpan.size() > 0 ? m.originalSpan.get(0) : sent.get(m.startIndex);
            m.headString = m.originalSpan.toString();
        }
    }
}