use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.
the class ArabicSegmenter method segmentStringToTokenList.
public List<CoreLabel> segmentStringToTokenList(String line) {
List<CoreLabel> tokenList = CollectionUtils.makeList();
List<CoreLabel> labeledSequence = segmentStringToIOB(line);
for (IntPair span : IOBUtils.TokenSpansForIOB(labeledSequence)) {
CoreLabel token = new CoreLabel();
String text = IOBUtils.IOBToString(labeledSequence, prefixMarker, suffixMarker, span.getSource(), span.getTarget());
token.setWord(text);
token.setValue(text);
token.set(CoreAnnotations.TextAnnotation.class, text);
token.set(CoreAnnotations.ArabicSegAnnotation.class, "1");
int start = labeledSequence.get(span.getSource()).beginPosition();
int end = labeledSequence.get(span.getTarget() - 1).endPosition();
token.setOriginalText(line.substring(start, end));
token.set(CoreAnnotations.CharacterOffsetBeginAnnotation.class, start);
token.set(CoreAnnotations.CharacterOffsetEndAnnotation.class, end);
tokenList.add(token);
}
return tokenList;
}
use of edu.stanford.nlp.util.IntPair 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);
}
}
}
}
}
use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.
the class DocumentPreprocessor method findSpeakersInArticle.
private static void findSpeakersInArticle(Document doc, Dictionaries dict) {
List<CoreMap> sentences = doc.annotation.get(CoreAnnotations.SentencesAnnotation.class);
IntPair beginQuotation = null;
IntPair endQuotation = null;
boolean insideQuotation = false;
int utterNum = -1;
for (int i = 0; i < sentences.size(); i++) {
List<CoreLabel> sent = sentences.get(i).get(CoreAnnotations.TokensAnnotation.class);
for (int j = 0; j < sent.size(); j++) {
int utterIndex = sent.get(j).get(CoreAnnotations.UtteranceAnnotation.class);
if (utterIndex != 0 && !insideQuotation) {
utterNum = utterIndex;
insideQuotation = true;
beginQuotation = new IntPair(i, j);
} else if (utterIndex == 0 && insideQuotation) {
insideQuotation = false;
endQuotation = new IntPair(i, j);
findQuotationSpeaker(doc, utterNum, sentences, beginQuotation, endQuotation, dict);
}
}
}
if (insideQuotation) {
endQuotation = new IntPair(sentences.size() - 1, sentences.get(sentences.size() - 1).get(CoreAnnotations.TokensAnnotation.class).size() - 1);
findQuotationSpeaker(doc, utterNum, sentences, beginQuotation, endQuotation, dict);
}
}
use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.
the class DocumentPreprocessor method findTwinMentionsRelaxed.
/** Mark twin mentions: heads of the mentions are matched */
private static void findTwinMentionsRelaxed(Document doc) {
for (int sentNum = 0; sentNum < doc.goldMentions.size(); sentNum++) {
List<Mention> golds = doc.goldMentions.get(sentNum);
List<Mention> predicts = doc.predictedMentions.get(sentNum);
Map<IntPair, Mention> goldMentionPositions = Generics.newHashMap();
Map<Integer, LinkedList<Mention>> goldMentionHeadPositions = Generics.newHashMap();
for (Mention g : golds) {
goldMentionPositions.put(new IntPair(g.startIndex, g.endIndex), g);
if (!goldMentionHeadPositions.containsKey(g.headIndex)) {
goldMentionHeadPositions.put(g.headIndex, new LinkedList<>());
}
goldMentionHeadPositions.get(g.headIndex).add(g);
}
List<Mention> remains = new ArrayList<>();
for (Mention p : predicts) {
IntPair pos = new IntPair(p.startIndex, p.endIndex);
if (goldMentionPositions.containsKey(pos)) {
Mention g = goldMentionPositions.get(pos);
p.mentionID = g.mentionID;
p.hasTwin = true;
g.hasTwin = true;
goldMentionHeadPositions.get(g.headIndex).remove(g);
if (goldMentionHeadPositions.get(g.headIndex).isEmpty()) {
goldMentionHeadPositions.remove(g.headIndex);
}
} else
remains.add(p);
}
for (Mention r : remains) {
if (goldMentionHeadPositions.containsKey(r.headIndex)) {
Mention g = goldMentionHeadPositions.get(r.headIndex).poll();
r.mentionID = g.mentionID;
r.hasTwin = true;
g.hasTwin = true;
if (goldMentionHeadPositions.get(g.headIndex).isEmpty()) {
goldMentionHeadPositions.remove(g.headIndex);
}
}
}
}
}
use of edu.stanford.nlp.util.IntPair in project CoreNLP by stanfordnlp.
the class DocumentPreprocessor method findTwinMentionsStrict.
/** Mark twin mentions: All mention boundaries should be matched */
private static void findTwinMentionsStrict(Document doc) {
for (int sentNum = 0; sentNum < doc.goldMentions.size(); sentNum++) {
List<Mention> golds = doc.goldMentions.get(sentNum);
List<Mention> predicts = doc.predictedMentions.get(sentNum);
// For CoNLL training there are some documents with gold mentions with the same position offsets
// See /scr/nlp/data/conll-2011/v2/data/train/data/english/annotations/nw/wsj/09/wsj_0990.v2_auto_conll
// (Packwood - Roth)
CollectionValuedMap<IntPair, Mention> goldMentionPositions = new CollectionValuedMap<>();
for (Mention g : golds) {
IntPair ip = new IntPair(g.startIndex, g.endIndex);
if (goldMentionPositions.containsKey(ip)) {
StringBuilder existingMentions = new StringBuilder();
for (Mention eg : goldMentionPositions.get(ip)) {
if (existingMentions.length() > 0) {
existingMentions.append(",");
}
existingMentions.append(eg.mentionID);
}
Redwood.log("debug-preprocessor", "WARNING: gold mentions with the same offsets: " + ip + " mentions=" + g.mentionID + "," + existingMentions + ", " + g.spanToString());
}
//assert(!goldMentionPositions.containsKey(ip));
goldMentionPositions.add(new IntPair(g.startIndex, g.endIndex), g);
}
for (Mention p : predicts) {
IntPair pos = new IntPair(p.startIndex, p.endIndex);
if (goldMentionPositions.containsKey(pos)) {
Collection<Mention> cm = goldMentionPositions.get(pos);
int minId = Integer.MAX_VALUE;
Mention g = null;
for (Mention m : cm) {
if (m.mentionID < minId) {
g = m;
minId = m.mentionID;
}
}
cm.remove(g);
p.mentionID = g.mentionID;
p.hasTwin = true;
g.hasTwin = true;
}
}
}
}
Aggregations