private void addMissingAnnotation(Annotation anno) {
boolean useConstituency = CorefProperties.useConstituencyTree(props);
final boolean LEMMATIZE = true;
List<CoreMap> sentences = anno.get(CoreAnnotations.SentencesAnnotation.class);
for (CoreMap sentence : sentences) {
boolean hasTree = sentence.containsKey(TreeCoreAnnotations.TreeAnnotation.class);
Tree tree = sentence.get(TreeCoreAnnotations.TreeAnnotation.class);
if (!useConstituency) { // TODO: temp for dev: make sure we don't use constituency tree
sentence.remove(TreeCoreAnnotations.TreeAnnotation.class);
}
if (LEMMATIZE && hasTree && useConstituency)
treeLemmatizer.transformTree(tree); // TODO don't need?
}
corenlp.annotate(anno);
}
@Override
public Document nextDoc() throws Exception {
List<List<CoreLabel>> allWords = new ArrayList<List<CoreLabel>>();
List<Tree> allTrees = new ArrayList<Tree>();
CoNLL2011DocumentReader.Document conllDoc = reader.getNextDocument();
if (conllDoc == null) {
return null;
}
Annotation anno = conllDoc.getAnnotation();
List<CoreMap> sentences = anno.get(CoreAnnotations.SentencesAnnotation.class);
for (CoreMap sentence : sentences) {
if (!Constants.USE_GOLD_PARSES && !replicateCoNLL) {
// Remove tree from annotation and replace with parse using stanford parser
sentence.remove(TreeCoreAnnotations.TreeAnnotation.class);
} else {
Tree tree = sentence.get(TreeCoreAnnotations.TreeAnnotation.class);
// generate the dependency graph
try {
SemanticGraph deps =
SemanticGraphFactory.makeFromTree(
tree, SemanticGraphFactory.Mode.COLLAPSED, includeExtras, lemmatize, threadSafe);
SemanticGraph basicDeps =
SemanticGraphFactory.makeFromTree(
tree, SemanticGraphFactory.Mode.BASIC, includeExtras, lemmatize, threadSafe);
sentence.set(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class, basicDeps);
sentence.set(SemanticGraphCoreAnnotations.CollapsedDependenciesAnnotation.class, deps);
} catch (Exception e) {
logger.log(
Level.WARNING,
"Exception caught during extraction of Stanford dependencies. Will ignore and continue...",
e);
}
}
}
String preSpeaker = null;
int utterance = -1;
for (CoreLabel token : anno.get(CoreAnnotations.TokensAnnotation.class)) {
if (!token.containsKey(CoreAnnotations.SpeakerAnnotation.class)) {
token.set(CoreAnnotations.SpeakerAnnotation.class, "");
}
String curSpeaker = token.get(CoreAnnotations.SpeakerAnnotation.class);
if (!curSpeaker.equals(preSpeaker)) {
utterance++;
preSpeaker = curSpeaker;
}
token.set(CoreAnnotations.UtteranceAnnotation.class, utterance);
}
// Run pipeline
stanfordProcessor.annotate(anno);
for (CoreMap sentence : anno.get(CoreAnnotations.SentencesAnnotation.class)) {
allWords.add(sentence.get(CoreAnnotations.TokensAnnotation.class));
allTrees.add(sentence.get(TreeCoreAnnotations.TreeAnnotation.class));
}
// Initialize gold mentions
List<List<Mention>> allGoldMentions = extractGoldMentions(conllDoc);
List<List<Mention>> allPredictedMentions;
if (Constants.USE_GOLD_MENTIONS) {
// allPredictedMentions = allGoldMentions;
// Make copy of gold mentions since mentions may be later merged, mentionID's changed and
// stuff
allPredictedMentions = makeCopy(allGoldMentions);
} else if (Constants.USE_GOLD_MENTION_BOUNDARIES) {
allPredictedMentions =
((RuleBasedCorefMentionFinder) mentionFinder)
.filterPredictedMentions(allGoldMentions, anno, dictionaries);
} else {
allPredictedMentions = mentionFinder.extractPredictedMentions(anno, maxID, dictionaries);
}
try {
recallErrors(allGoldMentions, allPredictedMentions, anno);
} catch (IOException e) {
throw new RuntimeException(e);
}
Document doc = arrange(anno, allWords, allTrees, allPredictedMentions, allGoldMentions, true);
doc.conllDoc = conllDoc;
return doc;
}