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);
}
/** Load Stanford Processor: skip unnecessary annotator */
protected StanfordCoreNLP loadStanfordProcessor(Properties props) {
Properties pipelineProps = new Properties(props);
StringBuilder annoSb = new StringBuilder("");
if (!CorefProperties.useGoldPOS(props)) {
annoSb.append("pos, lemma");
} else {
annoSb.append("lemma");
}
if (CorefProperties.USE_TRUECASE) {
annoSb.append(", truecase");
}
if (!CorefProperties.useGoldNE(props) || CorefProperties.getLanguage(props) == Locale.CHINESE) {
annoSb.append(", ner");
}
if (!CorefProperties.useGoldParse(props)) {
if (CorefProperties.useConstituencyTree(props)) annoSb.append(", parse");
else annoSb.append(", depparse");
}
String annoStr = annoSb.toString();
Redwood.log("MentionExtractor ignores specified annotators, using annotators=" + annoStr);
pipelineProps.put("annotators", annoStr);
return new StanfordCoreNLP(pipelineProps, false);
}
