in ctakes-coreference/src/main/java/org/apache/ctakes/coreference/ae/CoreferenceChainCoNLLWriter.java [38:257]
public void writeCas(JCas jCas) throws AnalysisEngineProcessException {
String myView = jCas.getViewName();
File filename = new File(ViewUriUtil.getURI(jCas));
// JCas chainsCas = null;
// try {
// chainsCas = goldViewName != null ? jCas.getView(goldViewName) : jCas;
// } catch (CASException e) {
// e.printStackTrace();
// throw new AnalysisEngineProcessException(e);
// }
int chainNum = 1;
HashMap<Annotation, Integer> ent2chain = new HashMap<>();
// if(isGold) System.out.println("\nGold chains:");
// else{
// icOut.println(String.format("#begin document (%s); part 000", filename.getPath()));
// System.out.println("\nChains:");
// }
Collection<CollectionTextRelation> rels = JCasUtil.select(jCas, CollectionTextRelation.class);
if(rels.size() == 0){
return;
}
// build a map from every markable that is in a chain to the chain number it is in (number is not important as long as they are
// distinct so we just number them in the order the uima annotation in gives them to us)
// This has to be reverse compatible with older coref module that added RelationArguments to a chain instead of Markables.
// So we grab the chain elements, check their type, then grab the markable annotation depending on that type.
for(CollectionTextRelation chain : rels){
FSList members = chain.getMembers();
// if we are doing cluster-mention coreference, some clusters will be singletons, we do not use those in conll scoring
if(members instanceof NonEmptyFSList &&
((NonEmptyFSList)members).getTail() instanceof EmptyFSList) continue;
while(members instanceof NonEmptyFSList){
TOP head = ((NonEmptyFSList) members).getHead();
Annotation mention = null;
if(head instanceof Annotation){
mention = (Annotation) head;
}else{
mention = ((RelationArgument)head).getArgument();
}
// Annotation mention = (Annotation) ((NonEmptyFSList) members).getHead();
ent2chain.put(mention, chainNum);
members = ((NonEmptyFSList)members).getTail();
System.out.print("Mention: " + mention.getCoveredText().replace("\n", "<CR>"));
System.out.print(" (" + mention.getBegin() + ", " + mention.getEnd() + ")");
if(!mention.getView().getViewName().equals(myView)){
System.out.print("[DOC:" + mention.getView().getViewName() + "]");
}
System.out.print(" -----> ");
}
System.out.println();
chainNum++;
}
// Here we are using newline tokens to delimit sentences because the sentence
// breaks that cTAKES creates may not be correct and some gold markables might
// wrap sentences which might be confusing to the consumer of this file.
out.println("#begin document (" + filename.getPath() + "); part 000");
List<BaseToken> tokens = new ArrayList<>(JCasUtil.select(jCas, BaseToken.class));
Multiset<Integer> endSet = HashMultiset.create();
int tokenId = 0;
int sentId = 0;
BaseToken nextToken = tokens.get(0);
for(int i = 0; i < tokens.size(); i++){
boolean endSentToken = false;
BaseToken token = nextToken;
if(i+1 < tokens.size()){
nextToken = tokens.get(i+1);
if(nextToken instanceof NewlineToken || (token.getCoveredText().equals(".") && !(endSet.size() > 0))){
endSentToken = true;
}
}
// if we see a newline token at the end of a sentence break the sentence
// only print out if we are not at the start of the sentence:
if(token instanceof NewlineToken){
continue;
}
int lastInd = token.getEnd();
// fix for some bad tokenization
if(token.getCoveredText().length() > 1 && token.getCoveredText().endsWith(".")){
lastInd = token.getEnd()-1;
}
List<Markable> markables = new ArrayList<>(JCasUtil.selectCovering(jCas, Markable.class, token.getBegin(), lastInd));
List<Annotation> startMention = new ArrayList<>();
Multiset<Integer> endMention = HashMultiset.create();
List<Integer> wholeMention = new ArrayList<>();
for(Annotation markable : markables){
if(ent2chain.containsKey(markable)){
if(markable.getBegin() == token.getBegin()){
if(markable.getEnd() == token.getEnd()){
wholeMention.add(ent2chain.get(markable));
}else{
startMention.add(markable);
}
}else if(markable.getEnd() <= token.getEnd()){
if(endMention.contains(ent2chain.get(markable))){
System.err.println("There is a duplicate element -- should be handled by multiset");
}
if(markable.getEnd() < token.getEnd()){
System.err.println("There is a markable that ends in the middle of a token!");
}
endMention.add(ent2chain.get(markable));
}
// if(!isGold){
// icOut.println(String.format("%d-%d-%d\n", sentId, markable.getBegin(), markable.getEnd()));
// }
}
}
out.print(filename.getPath());
out.print('\t');
out.print(docNum);
out.print('\t');
out.print(tokenId++);
out.print('\t');
out.print(token instanceof NewlineToken ? "Newline" : TreeUtils.escapePunct(token.getCoveredText()));
out.print('\t');
out.print(token.getPartOfSpeech());
out.print('\t');
// parse bit -- assume flat parse
if(tokenId == 1){
out.print("(NOPARSE*");
// special case for one word sentences:
if(endSentToken){
out.print(")");
}
}else if(endSentToken){
out.print("*)");
}else{
out.print("*");
}
out.print('\t');
// predicate lemma -- can ignore?
out.print('-'); out.print('\t');
// predicate frameset id -- can ignore?
out.print('-'); out.print('\t');
// word sense
out.print('-'); out.print('\t');
// speaker/author
out.print('-'); out.print('\t');
// named entities
out.print('*'); out.print('\t');
StringBuffer buff = new StringBuffer();
// while(endStack.size() > 0 && endMention.contains(endStack.peek())){
for(int ind : endMention){
// int ind = endStack.pop();
// int ind = endMention.get(j);
if(endSet.contains(ind)){
buff.append(ind);
buff.append(')');
buff.append('|');
}
endSet.remove(ind);
// endMention.remove(ind);
}
for(int ind : wholeMention){
buff.append('(');
buff.append(ind);
buff.append(')');
buff.append('|');
}
// sort start mentions by ordering of ending
while(startMention.size() > 0){
int ind;
Annotation latestEnd = null;
for(int j = 0; j < startMention.size(); j++){
if(latestEnd == null || startMention.get(j).getEnd() > latestEnd.getEnd()){
latestEnd = startMention.get(j);
}
}
startMention.remove(latestEnd);
ind = ent2chain.get(latestEnd);
buff.append('(');
buff.append(ind);
buff.append('|');
endSet.add(ind);
// endStack.push(ind);
}
// In some datasets markables end in the middle of a token -- this is a problem because our check above is for all markables that cover the
// current token. In this case the markable end will still be unused when we get to the end of the sentence. We'll just hack it by throwing
// them on the last token of the sentence.
if(endSentToken && endSet.size() > 0){
System.err.println("Error! There are opened markables that never closed! Putting them on the end of the sentence.");
for(int ind : endSet){
buff.append(ind);
buff.append(')');
buff.append('|');
}
endSet.clear();
}
if(buff.length() > 0){
out.println(buff.substring(0, buff.length()-1));
}else{
out.println("_");
}
if(endSentToken){
out.println();
tokenId = 0;
sentId++;
}
}
// if(!isGold){
// icOut.println("#end document");
// }
out.println("#end document " + filename.getPath());
out.flush();
docNum++;
}