use of de.tudarmstadt.ukp.dkpro.core.api.coref.type.CoreferenceChain in project webanno by webanno.
the class TcfReader method convertCoreference.
/**
* Correferences in CAS should be represented {@link CoreferenceChain} and
* {@link CoreferenceLink}. The TCF representation Uses <b> rel </b> and
* <b>target </b> to build chains. Example: </br>
* <i> {@literal <entity><reference ID="rc_0" tokenIDs="t_0" mintokIDs=
* "t_0" type="nam"/> } </br>
* {@literal <reference ID="rc_1" tokenIDs="t_6" mintokIDs="t_6" type=
* "pro.per3" rel=
* "anaphoric" target="rc_0"/></entity>
* }</i> </br>
* The first phase of conversion is getting all <b>references</b> and
* <b>targets</b> alongside the <b>type</b> and <b>relations in different
* maps</b> <br>
* Second, an iteration is made through all the maps and the
* {@link CoreferenceChain} and {@link CoreferenceLink} annotations are
* constructed.
*/
private void convertCoreference(JCas aJCas, TextCorpus aCorpusData, Map<String, Token> aTokens) {
if (aCorpusData.getReferencesLayer() == null) {
// No layer to read from.
return;
}
for (int i = 0; i < aCorpusData.getReferencesLayer().size(); i++) {
eu.clarin.weblicht.wlfxb.tc.api.ReferencedEntity entity = aCorpusData.getReferencesLayer().getReferencedEntity(i);
Map<Integer, CoreferenceLink> referencesMap = new TreeMap<>();
storeReferencesAndTargetsInMap(referencesMap, entity, aCorpusData, aTokens, aJCas);
CoreferenceChain chain = new CoreferenceChain(aJCas);
CoreferenceLink link = null;
for (Integer address : referencesMap.keySet()) {
if (chain.getFirst() == null) {
chain.setFirst(referencesMap.get(address));
link = chain.getFirst();
chain.addToIndexes();
} else {
link.setNext(referencesMap.get(address));
if (link.getReferenceRelation() == null) {
link.setReferenceRelation(referencesMap.get(address).getReferenceRelation());
}
link = link.getNext();
link.addToIndexes();
}
}
}
}
use of de.tudarmstadt.ukp.dkpro.core.api.coref.type.CoreferenceChain in project webanno by webanno.
the class TcfWriter method writeCoreference.
private void writeCoreference(JCas aJCas, TextCorpus aTextCorpus, Map<Integer, eu.clarin.weblicht.wlfxb.tc.api.Token> aTokensBeginPositionMap) {
if (!JCasUtil.exists(aJCas, CoreferenceChain.class)) {
// Do nothing if there are no coreference chains in the CAS
getLogger().debug("Layer [" + TextCorpusLayerTag.REFERENCES.getXmlName() + "]: empty");
return;
}
String tagSetName = "TueBaDz";
for (TagsetDescription tagSet : select(aJCas, TagsetDescription.class)) {
if (tagSet.getLayer().equals(CoreferenceLink.class.getName())) {
tagSetName = tagSet.getName();
break;
}
}
ReferencesLayer coreferencesLayer = aTextCorpus.createReferencesLayer(null, tagSetName, null);
getLogger().debug("Layer [" + TextCorpusLayerTag.REFERENCES.getXmlName() + "]: created");
for (CoreferenceChain chain : select(aJCas, CoreferenceChain.class)) {
CoreferenceLink prevLink = null;
Reference prevRef = null;
List<Reference> refs = new ArrayList<>();
for (CoreferenceLink link : chain.links()) {
// Get covered tokens
List<eu.clarin.weblicht.wlfxb.tc.api.Token> tokens = new ArrayList<>();
for (Token token : selectCovered(Token.class, link)) {
tokens.add(aTokensBeginPositionMap.get(token.getBegin()));
}
// Create current reference
Reference ref = coreferencesLayer.createReference(link.getReferenceType(), tokens, null);
// Special handling for expletive relations
if (REL_TYPE_EXPLETIVE.equals(link.getReferenceRelation())) {
coreferencesLayer.addRelation(ref, REL_TYPE_EXPLETIVE);
// chain, so we bail out here.
continue;
}
// Create relation between previous and current reference
if (prevLink != null) {
coreferencesLayer.addRelation(prevRef, prevLink.getReferenceRelation(), ref);
}
prevLink = link;
prevRef = ref;
refs.add(ref);
}
coreferencesLayer.addReferent(refs);
}
}
use of de.tudarmstadt.ukp.dkpro.core.api.coref.type.CoreferenceChain in project webanno by webanno.
the class WebAnnoSemanticGraphReader method convertToCas.
public void convertToCas(JCas aJCas, InputStream aIs, String aEncoding) throws IOException {
StringBuilder text = new StringBuilder();
LineIterator lineIterator = IOUtils.lineIterator(aIs, aEncoding);
int tokenBeginPosition = 0;
while (lineIterator.hasNext()) {
String line = lineIterator.next();
String[] contents = line.split("\t>\t|\tX\t");
int sentenceBegin = tokenBeginPosition;
int chainBegin = tokenBeginPosition;
int chainEnd = 0;
StringTokenizer st = new StringTokenizer(contents[0]);
while (st.hasMoreTokens()) {
String content = st.nextToken();
Token outToken = new Token(aJCas, tokenBeginPosition, tokenBeginPosition + content.length());
outToken.addToIndexes();
tokenBeginPosition = outToken.getEnd() + 1;
chainEnd = tokenBeginPosition;
text.append(content).append(" ");
}
CoreferenceChain chain = new CoreferenceChain(aJCas);
CoreferenceLink link = new CoreferenceLink(aJCas, chainBegin, chainEnd - 1);
link.setReferenceType("text");
link.addToIndexes();
chain.setFirst(link);
if (line.contains("\t>\t")) {
link.setReferenceRelation("entails");
Token outToken = new Token(aJCas, tokenBeginPosition, tokenBeginPosition + 1);
outToken.addToIndexes();
tokenBeginPosition = outToken.getEnd() + 1;
text.append("> ");
} else {
link.setReferenceRelation("do not entails");
Token outToken = new Token(aJCas, tokenBeginPosition, tokenBeginPosition + 1);
outToken.addToIndexes();
tokenBeginPosition = outToken.getEnd() + 1;
text.append("X ");
}
chainBegin = tokenBeginPosition;
st = new StringTokenizer(contents[0]);
while (st.hasMoreTokens()) {
String content = st.nextToken();
Token outToken = new Token(aJCas, tokenBeginPosition, tokenBeginPosition + content.length());
outToken.addToIndexes();
tokenBeginPosition = outToken.getEnd() + 1;
chainEnd = tokenBeginPosition;
text.append(content).append(" ");
}
CoreferenceLink nextLink = new CoreferenceLink(aJCas, chainBegin, chainEnd - 1);
nextLink.setReferenceType("hypothesis");
nextLink.addToIndexes();
link.setNext(nextLink);
chain.addToIndexes();
text.append("\n");
Sentence outSentence = new Sentence(aJCas);
outSentence.setBegin(sentenceBegin);
outSentence.setEnd(tokenBeginPosition);
outSentence.addToIndexes();
tokenBeginPosition = tokenBeginPosition + 1;
sentenceBegin = tokenBeginPosition;
}
aJCas.setDocumentText(text.toString());
}
Aggregations