public static void fillInParseAnnotations(
boolean verbose, boolean buildGraphs, CoreMap sentence, Tree tree) {
// make sure all tree nodes are CoreLabels
// TODO: why isn't this always true? something fishy is going on
ParserAnnotatorUtils.convertToCoreLabels(tree);
// index nodes, i.e., add start and end token positions to all nodes
// this is needed by other annotators down stream, e.g., the NFLAnnotator
tree.indexSpans(0);
sentence.set(TreeAnnotation.class, tree);
if (verbose) {
System.err.println("Tree is:");
tree.pennPrint(System.err);
}
if (buildGraphs) {
// generate the dependency graph
SemanticGraph deps = generateCollapsedDependencies(tree);
SemanticGraph uncollapsedDeps = generateUncollapsedDependencies(tree);
SemanticGraph ccDeps = generateCCProcessedDependencies(tree);
if (verbose) {
System.err.println("SDs:");
System.err.println(deps.toString("plain"));
}
sentence.set(SemanticGraphCoreAnnotations.CollapsedDependenciesAnnotation.class, deps);
sentence.set(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation.class, uncollapsedDeps);
sentence.set(
SemanticGraphCoreAnnotations.CollapsedCCProcessedDependenciesAnnotation.class, ccDeps);
}
setMissingTags(sentence, tree);
}
public static void addFigerAnnotationToDocument(Annotation d) throws SQLException {
List<CoreMap> sentences = d.get(CoreAnnotations.SentencesAnnotation.class);
Set<String> entityIds = new HashSet<String>();
for (CoreMap sen : sentences) {
List<Triple<Pair<Integer, Integer>, String, Float>> nelAnnotation =
sen.get(NamedEntityLinkingAnnotation.class);
for (Triple<Pair<Integer, Integer>, String, Float> t : nelAnnotation) {
String id = t.second;
if (!id.equals("null")) {
entityIds.add(id);
}
}
}
Map<String, Set<String>> idTypeMap = bigQuery(entityIds);
// add type onto sentences
for (CoreMap sen : sentences) {
List<Triple<Pair<Integer, Integer>, String, Float>> nelAnnotation =
sen.get(NamedEntityLinkingAnnotation.class);
List<Triple<Set<String>, Integer, Integer>> figerData = new ArrayList<>();
for (Triple<Pair<Integer, Integer>, String, Float> t : nelAnnotation) {
Integer start = t.first.first;
Integer end = t.first.second;
Set<String> types = null;
if (!t.second.equals("null")) {
types = idTypeMap.get(GuidMidConversion.convertBackward(t.second));
}
Triple<Set<String>, Integer, Integer> figerTrip = new Triple<>(types, start, end);
figerData.add(figerTrip);
}
sen.set(FigerAnnotation.class, figerData);
}
}
private static void addLemma(
Morphology morpha,
Class<? extends CoreAnnotation<String>> ann,
CoreMap map,
String word,
String tag) {
if (tag.length() > 0) {
String phrasalVerb = phrasalVerb(morpha, word, tag);
if (phrasalVerb == null) {
map.set(ann, morpha.lemma(word, tag));
} else {
map.set(ann, phrasalVerb);
}
} else {
map.set(ann, morpha.stem(word));
}
}
private List<CoreMap> toCoreMaps(
CoreMap annotation, List<TimeExpression> timeExpressions, SUTime.TimeIndex timeIndex) {
if (timeExpressions == null) return null;
List<CoreMap> coreMaps = new ArrayList<CoreMap>(timeExpressions.size());
for (TimeExpression te : timeExpressions) {
CoreMap cm = te.getAnnotation();
SUTime.Temporal temporal = te.getTemporal();
if (temporal != null) {
String origText = annotation.get(CoreAnnotations.TextAnnotation.class);
String text = cm.get(CoreAnnotations.TextAnnotation.class);
if (origText != null) {
// Make sure the text is from original (and not from concatenated tokens)
ChunkAnnotationUtils.annotateChunkText(cm, annotation);
text = cm.get(CoreAnnotations.TextAnnotation.class);
}
Map<String, String> timexAttributes;
try {
timexAttributes = temporal.getTimexAttributes(timeIndex);
if (options.includeRange) {
SUTime.Temporal rangeTemporal = temporal.getRange();
if (rangeTemporal != null) {
timexAttributes.put("range", rangeTemporal.toString());
}
}
} catch (Exception e) {
logger.log(
Level.WARNING,
"Failed to get attributes from " + text + ", timeIndex " + timeIndex,
e);
continue;
}
Timex timex;
try {
timex = Timex.fromMap(text, timexAttributes);
} catch (Exception e) {
logger.log(
Level.WARNING,
"Failed to process " + text + " with attributes " + timexAttributes,
e);
continue;
}
cm.set(TimexAnnotation.class, timex);
if (timex != null) {
coreMaps.add(cm);
} else {
logger.warning("No timex expression for: " + text);
}
}
}
return coreMaps;
}
public List<TimeExpression> extractTimeExpressions(CoreMap annotation, String docDateStr) {
List<CoreMap> mergedNumbers = NumberNormalizer.findAndMergeNumbers(annotation);
annotation.set(CoreAnnotations.NumerizedTokensAnnotation.class, mergedNumbers);
// TODO: docDate may not have century....
SUTime.Time docDate = timexPatterns.parseDateTime(docDateStr);
List<? extends MatchedExpression> matchedExpressions =
expressionExtractor.extractExpressions(annotation);
List<TimeExpression> timeExpressions = new ArrayList<TimeExpression>(matchedExpressions.size());
for (MatchedExpression expr : matchedExpressions) {
if (expr instanceof TimeExpression) {
timeExpressions.add((TimeExpression) expr);
} else {
timeExpressions.add(new TimeExpression(expr));
}
}
// Add back nested time expressions for ranges....
// For now only one level of nesting...
if (options.includeNested) {
List<TimeExpression> nestedTimeExpressions = new ArrayList<TimeExpression>();
for (TimeExpression te : timeExpressions) {
if (te.isIncludeNested()) {
List<? extends CoreMap> children =
te.getAnnotation().get(TimeExpression.ChildrenAnnotation.class);
if (children != null) {
for (CoreMap child : children) {
TimeExpression childTe = child.get(TimeExpression.Annotation.class);
if (childTe != null) {
nestedTimeExpressions.add(childTe);
}
}
}
}
}
timeExpressions.addAll(nestedTimeExpressions);
}
Collections.sort(timeExpressions, MatchedExpression.EXPR_TOKEN_OFFSETS_NESTED_FIRST_COMPARATOR);
timeExpressions = filterInvalidTimeExpressions(timeExpressions);
// Some resolving is done even if docDate null...
if (
/*docDate != null && */ timeExpressions != null) {
resolveTimeExpressions(annotation, timeExpressions, docDate);
}
// Annotate timex
return timeExpressions;
}
/** Reads an annotation from the given filename using the requested input. */
public static List<Annotation> getAnnotations(
StanfordCoreNLP tokenizer, Input inputFormat, String filename, boolean filterUnknown) {
switch (inputFormat) {
case TEXT:
{
String text = IOUtils.slurpFileNoExceptions(filename);
Annotation annotation = new Annotation(text);
tokenizer.annotate(annotation);
List<Annotation> annotations = Generics.newArrayList();
for (CoreMap sentence : annotation.get(CoreAnnotations.SentencesAnnotation.class)) {
Annotation nextAnnotation =
new Annotation(sentence.get(CoreAnnotations.TextAnnotation.class));
nextAnnotation.set(
CoreAnnotations.SentencesAnnotation.class, Collections.singletonList(sentence));
annotations.add(nextAnnotation);
}
return annotations;
}
case TREES:
{
List<Tree> trees;
if (filterUnknown) {
trees = SentimentUtils.readTreesWithGoldLabels(filename);
trees = SentimentUtils.filterUnknownRoots(trees);
} else {
trees = Generics.newArrayList();
MemoryTreebank treebank = new MemoryTreebank("utf-8");
treebank.loadPath(filename, null);
for (Tree tree : treebank) {
trees.add(tree);
}
}
List<Annotation> annotations = Generics.newArrayList();
for (Tree tree : trees) {
CoreMap sentence = new Annotation(Sentence.listToString(tree.yield()));
sentence.set(TreeCoreAnnotations.TreeAnnotation.class, tree);
List<CoreMap> sentences = Collections.singletonList(sentence);
Annotation annotation = new Annotation("");
annotation.set(CoreAnnotations.SentencesAnnotation.class, sentences);
annotations.add(annotation);
}
return annotations;
}
default:
throw new IllegalArgumentException("Unknown format " + inputFormat);
}
}
/**
* Converts NamedEntityTagAnnotation tags into {@link EntityMention}s. This finds the longest
* sequence of NamedEntityTagAnnotation tags of the matching type.
*
* @param sentence A sentence annotated with NamedEntityTagAnnotation
*/
public void makeAnnotationFromAllNERTags(CoreMap sentence) {
List<CoreLabel> words = sentence.get(CoreAnnotations.TokensAnnotation.class);
List<EntityMention> mentions =
sentence.get(MachineReadingAnnotations.EntityMentionsAnnotation.class);
assert words != null;
if (mentions == null) {
this.logger.info("mentions are null");
mentions = new ArrayList<>();
}
for (int start = 0; start < words.size(); start++) {
int end;
// find the first token after start that isn't of nerType
String lastneTag = null;
String ne = null;
for (end = start; end < words.size(); end++) {
ne = words.get(end).get(NamedEntityTagAnnotation.class);
if (ne.equals(SeqClassifierFlags.DEFAULT_BACKGROUND_SYMBOL)
|| (lastneTag != null && !ne.equals(lastneTag))) {
break;
}
lastneTag = ne;
}
if (end > start) {
// found a match!
String entityType = this.getEntityTypeForTag(lastneTag);
EntityMention m =
EntityMentionFactory.constructEntityMention(
EntityMention.makeUniqueId(),
sentence,
new Span(start, end),
new Span(start, end),
entityType,
null,
null);
// TODO: changed entityType in the above sentence to nerTag - Sonal
logger.info("Created " + entityType + " entity mention: " + m);
start = end - 1;
mentions.add(m);
}
}
sentence.set(MachineReadingAnnotations.EntityMentionsAnnotation.class, mentions);
}
/**
* Converts NamedEntityTagAnnotation tags into {@link EntityMention}s. This finds the longest
* sequence of NamedEntityTagAnnotation tags of the matching type.
*
* @param sentence A sentence, ideally annotated with NamedEntityTagAnnotation
* @param nerTag The name of the NER tag to copy, e.g. "DATE".
* @param entityType The type of the {@link EntityMention} objects created
*/
public static void makeAnnotationFromGivenNERTag(
CoreMap sentence, String nerTag, String entityType) {
List<CoreLabel> words = sentence.get(CoreAnnotations.TokensAnnotation.class);
List<EntityMention> mentions =
sentence.get(MachineReadingAnnotations.EntityMentionsAnnotation.class);
assert words != null;
assert mentions != null;
for (int start = 0; start < words.size(); start++) {
int end;
// find the first token after start that isn't of nerType
for (end = start; end < words.size(); end++) {
String ne = words.get(end).get(NamedEntityTagAnnotation.class);
if (!ne.equals(nerTag)) {
break;
}
}
if (end > start) {
// found a match!
EntityMention m =
EntityMentionFactory.constructEntityMention(
EntityMention.makeUniqueId(),
sentence,
new Span(start, end),
new Span(start, end),
entityType,
null,
null);
logger.info("Created " + entityType + " entity mention: " + m);
start = end - 1;
mentions.add(m);
}
}
sentence.set(MachineReadingAnnotations.EntityMentionsAnnotation.class, mentions);
}
@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;
}
@Override
public Document nextDoc() throws Exception {
List<List<CoreLabel>> allWords = new ArrayList<List<CoreLabel>>();
List<Tree> allTrees = new ArrayList<Tree>();
List<List<Mention>> allGoldMentions = new ArrayList<List<Mention>>();
List<List<Mention>> allPredictedMentions;
List<CoreMap> allSentences = new ArrayList<CoreMap>();
Annotation docAnno = new Annotation("");
Pattern docPattern =
Pattern.compile("<DOC>(.*?)</DOC>", Pattern.DOTALL + Pattern.CASE_INSENSITIVE);
Pattern sentencePattern =
Pattern.compile(
"(<s>|<hl>|<dd>|<DATELINE>)(.*?)(</s>|</hl>|</dd>|</DATELINE>)",
Pattern.DOTALL + Pattern.CASE_INSENSITIVE);
Matcher docMatcher = docPattern.matcher(fileContents);
if (!docMatcher.find(currentOffset)) return null;
currentOffset = docMatcher.end();
String doc = docMatcher.group(1);
Matcher sentenceMatcher = sentencePattern.matcher(doc);
String ner = null;
// Maintain current document ID.
Pattern docIDPattern =
Pattern.compile("<DOCNO>(.*?)</DOCNO>", Pattern.DOTALL + Pattern.CASE_INSENSITIVE);
Matcher docIDMatcher = docIDPattern.matcher(doc);
if (docIDMatcher.find()) currentDocumentID = docIDMatcher.group(1);
else currentDocumentID = "documentAfter " + currentDocumentID;
while (sentenceMatcher.find()) {
String sentenceString = sentenceMatcher.group(2);
List<CoreLabel> words =
tokenizerFactory.getTokenizer(new StringReader(sentenceString)).tokenize();
// FIXING TOKENIZATION PROBLEMS
for (int i = 0; i < words.size(); i++) {
CoreLabel w = words.get(i);
if (i > 0 && w.word().equals("$")) {
if (!words.get(i - 1).word().endsWith("PRP") && !words.get(i - 1).word().endsWith("WP"))
continue;
words.get(i - 1).set(CoreAnnotations.TextAnnotation.class, words.get(i - 1).word() + "$");
words.remove(i);
i--;
} else if (w.word().equals("\\/")) {
if (words.get(i - 1).word().equals("</COREF>")) continue;
w.set(
CoreAnnotations.TextAnnotation.class,
words.get(i - 1).word() + "\\/" + words.get(i + 1).word());
words.remove(i + 1);
words.remove(i - 1);
}
}
// END FIXING TOKENIZATION PROBLEMS
List<CoreLabel> sentence = new ArrayList<CoreLabel>();
// MUC accepts embedded coref mentions, so we need to keep a stack for the mentions currently
// open
Stack<Mention> stack = new Stack<Mention>();
List<Mention> mentions = new ArrayList<Mention>();
allWords.add(sentence);
allGoldMentions.add(mentions);
for (CoreLabel word : words) {
String w = word.get(CoreAnnotations.TextAnnotation.class);
// found regular token: WORD/POS
if (!w.startsWith("<") && w.contains("\\/") && w.lastIndexOf("\\/") != w.length() - 2) {
int i = w.lastIndexOf("\\/");
String w1 = w.substring(0, i);
// we do NOT set POS info here. We take the POS tags from the parser!
word.set(CoreAnnotations.TextAnnotation.class, w1);
word.remove(CoreAnnotations.OriginalTextAnnotation.class);
if (Constants.USE_GOLD_NE) {
if (ner != null) {
word.set(CoreAnnotations.NamedEntityTagAnnotation.class, ner);
} else {
word.set(CoreAnnotations.NamedEntityTagAnnotation.class, "O");
}
}
sentence.add(word);
}
// found the start SGML tag for a NE, e.g., "<ORGANIZATION>"
else if (w.startsWith("<") && !w.startsWith("<COREF") && !w.startsWith("</")) {
Pattern nerPattern = Pattern.compile("<(.*?)>");
Matcher m = nerPattern.matcher(w);
m.find();
ner = m.group(1);
}
// found the end SGML tag for a NE, e.g., "</ORGANIZATION>"
else if (w.startsWith("</") && !w.startsWith("</COREF")) {
Pattern nerPattern = Pattern.compile("</(.*?)>");
Matcher m = nerPattern.matcher(w);
m.find();
String ner1 = m.group(1);
if (ner != null && !ner.equals(ner1))
throw new RuntimeException("Unmatched NE labels in MUC file: " + ner + " v. " + ner1);
ner = null;
}
// found the start SGML tag for a coref mention
else if (w.startsWith("<COREF")) {
Mention mention = new Mention();
// position of this mention in the sentence
mention.startIndex = sentence.size();
// extract GOLD info about this coref chain. needed for eval
Pattern idPattern = Pattern.compile("ID=\"(.*?)\"");
Pattern refPattern = Pattern.compile("REF=\"(.*?)\"");
Matcher m = idPattern.matcher(w);
m.find();
mention.mentionID = Integer.parseInt(m.group(1));
m = refPattern.matcher(w);
if (m.find()) {
mention.originalRef = Integer.parseInt(m.group(1));
}
// open mention. keep track of all open mentions using the stack
stack.push(mention);
}
// found the end SGML tag for a coref mention
else if (w.equals("</COREF>")) {
Mention mention = stack.pop();
mention.endIndex = sentence.size();
// this is a closed mention. add it to the final list of mentions
// System.err.printf("Found MENTION: ID=%d, REF=%d\n", mention.mentionID,
// mention.originalRef);
mentions.add(mention);
} else {
word.remove(CoreAnnotations.OriginalTextAnnotation.class);
if (Constants.USE_GOLD_NE) {
if (ner != null) {
word.set(CoreAnnotations.NamedEntityTagAnnotation.class, ner);
} else {
word.set(CoreAnnotations.NamedEntityTagAnnotation.class, "O");
}
}
sentence.add(word);
}
}
StringBuilder textContent = new StringBuilder();
for (int i = 0; i < sentence.size(); i++) {
CoreLabel w = sentence.get(i);
w.set(CoreAnnotations.IndexAnnotation.class, i + 1);
w.set(CoreAnnotations.UtteranceAnnotation.class, 0);
if (i > 0) textContent.append(" ");
textContent.append(w.getString(CoreAnnotations.TextAnnotation.class));
}
CoreMap sentCoreMap = new Annotation(textContent.toString());
allSentences.add(sentCoreMap);
sentCoreMap.set(CoreAnnotations.TokensAnnotation.class, sentence);
}
// assign goldCorefClusterID
Map<Integer, Mention> idMention = Generics.newHashMap(); // temporary use
for (List<Mention> goldMentions : allGoldMentions) {
for (Mention m : goldMentions) {
idMention.put(m.mentionID, m);
}
}
for (List<Mention> goldMentions : allGoldMentions) {
for (Mention m : goldMentions) {
if (m.goldCorefClusterID == -1) {
if (m.originalRef == -1) m.goldCorefClusterID = m.mentionID;
else {
int ref = m.originalRef;
while (true) {
Mention m2 = idMention.get(ref);
if (m2.goldCorefClusterID != -1) {
m.goldCorefClusterID = m2.goldCorefClusterID;
break;
} else if (m2.originalRef == -1) {
m2.goldCorefClusterID = m2.mentionID;
m.goldCorefClusterID = m2.goldCorefClusterID;
break;
} else {
ref = m2.originalRef;
}
}
}
}
}
}
docAnno.set(CoreAnnotations.SentencesAnnotation.class, allSentences);
stanfordProcessor.annotate(docAnno);
if (allSentences.size() != allWords.size())
throw new IllegalStateException("allSentences != allWords");
for (int i = 0; i < allSentences.size(); i++) {
List<CoreLabel> annotatedSent =
allSentences.get(i).get(CoreAnnotations.TokensAnnotation.class);
List<CoreLabel> unannotatedSent = allWords.get(i);
List<Mention> mentionInSent = allGoldMentions.get(i);
for (Mention m : mentionInSent) {
m.dependency =
allSentences
.get(i)
.get(SemanticGraphCoreAnnotations.CollapsedDependenciesAnnotation.class);
}
if (annotatedSent.size() != unannotatedSent.size()) {
throw new IllegalStateException("annotatedSent != unannotatedSent");
}
for (int j = 0, sz = annotatedSent.size(); j < sz; j++) {
CoreLabel annotatedWord = annotatedSent.get(j);
CoreLabel unannotatedWord = unannotatedSent.get(j);
if (!annotatedWord
.get(CoreAnnotations.TextAnnotation.class)
.equals(unannotatedWord.get(CoreAnnotations.TextAnnotation.class))) {
throw new IllegalStateException("annotatedWord != unannotatedWord");
}
}
allWords.set(i, annotatedSent);
allTrees.add(allSentences.get(i).get(TreeCoreAnnotations.TreeAnnotation.class));
}
// extract predicted mentions
if (Constants.USE_GOLD_MENTIONS) allPredictedMentions = allGoldMentions;
else
allPredictedMentions = mentionFinder.extractPredictedMentions(docAnno, maxID, dictionaries);
// add the relevant fields to mentions and order them for coref
return arrange(docAnno, allWords, allTrees, allPredictedMentions, allGoldMentions, true);
}
public void annotate(CoreMap document) throws IOException {
// write input file in GUTime format
Element inputXML = toInputXML(document);
File inputFile = File.createTempFile("gutime", ".input");
// Document doc = new Document(inputXML);
PrintWriter inputWriter = new PrintWriter(inputFile);
inputWriter.println(inputXML.toXML());
// new XMLOutputter().output(inputXML, inputWriter);
inputWriter.close();
boolean useFirstDate =
(!document.has(CoreAnnotations.CalendarAnnotation.class)
&& !document.has(CoreAnnotations.DocDateAnnotation.class));
ArrayList<String> args = new ArrayList<String>();
args.add("perl");
args.add("-I" + this.gutimePath.getPath());
args.add(new File(this.gutimePath, "TimeTag.pl").getPath());
if (useFirstDate) args.add("-FDNW");
args.add(inputFile.getPath());
// run GUTime on the input file
ProcessBuilder process = new ProcessBuilder(args);
StringWriter outputWriter = new StringWriter();
SystemUtils.run(process, outputWriter, null);
String output = outputWriter.getBuffer().toString();
Pattern docClose = Pattern.compile("</DOC>.*", Pattern.DOTALL);
output = docClose.matcher(output).replaceAll("</DOC>");
// parse the GUTime output
Element outputXML;
try {
Document newNodeDocument = new Builder().build(output, "");
outputXML = newNodeDocument.getRootElement();
} catch (ParsingException ex) {
throw new RuntimeException(
String.format(
"error:\n%s\ninput:\n%s\noutput:\n%s", ex, IOUtils.slurpFile(inputFile), output));
}
/*
try {
outputXML = new SAXBuilder().build(new StringReader(output)).getRootElement();
} catch (JDOMException e) {
throw new RuntimeException(String.format("error:\n%s\ninput:\n%s\noutput:\n%s",
e, IOUtils.slurpFile(inputFile), output));
} */
inputFile.delete();
// get Timex annotations
List<CoreMap> timexAnns = toTimexCoreMaps(outputXML, document);
document.set(TimexAnnotations.class, timexAnns);
if (outputResults) {
System.out.println(timexAnns);
}
// align Timex annotations to sentences
int timexIndex = 0;
for (CoreMap sentence : document.get(CoreAnnotations.SentencesAnnotation.class)) {
int sentBegin = beginOffset(sentence);
int sentEnd = endOffset(sentence);
// skip times before the sentence
while (timexIndex < timexAnns.size() && beginOffset(timexAnns.get(timexIndex)) < sentBegin) {
++timexIndex;
}
// determine times within the sentence
int sublistBegin = timexIndex;
int sublistEnd = timexIndex;
while (timexIndex < timexAnns.size()
&& sentBegin <= beginOffset(timexAnns.get(timexIndex))
&& endOffset(timexAnns.get(timexIndex)) <= sentEnd) {
++sublistEnd;
++timexIndex;
}
// set the sentence timexes
sentence.set(TimexAnnotations.class, timexAnns.subList(sublistBegin, sublistEnd));
}
}
private static List<CoreMap> toTimexCoreMaps(Element docElem, CoreMap originalDocument) {
// --Collect Token Offsets
HashMap<Integer, Integer> beginMap = new HashMap<Integer, Integer>();
HashMap<Integer, Integer> endMap = new HashMap<Integer, Integer>();
boolean haveTokenOffsets = true;
for (CoreMap sent : originalDocument.get(CoreAnnotations.SentencesAnnotation.class)) {
for (CoreLabel token : sent.get(CoreAnnotations.TokensAnnotation.class)) {
Integer tokBegin = token.get(CoreAnnotations.TokenBeginAnnotation.class);
Integer tokEnd = token.get(CoreAnnotations.TokenEndAnnotation.class);
if (tokBegin == null || tokEnd == null) {
haveTokenOffsets = false;
}
int charBegin = token.get(CoreAnnotations.CharacterOffsetBeginAnnotation.class);
int charEnd = token.get(CoreAnnotations.CharacterOffsetEndAnnotation.class);
beginMap.put(charBegin, tokBegin);
endMap.put(charEnd, tokEnd);
}
}
// --Set Timexes
List<CoreMap> timexMaps = new ArrayList<CoreMap>();
int offset = 0;
Element textElem = docElem.getFirstChildElement("text");
for (int i = 0; i < textElem.getChildCount(); i++) {
Node content = textElem.getChild(i);
if (content instanceof Text) {
Text text = (Text) content;
offset += text.getValue().length();
} else if (content instanceof Element) {
Element child = (Element) content;
if (child.getLocalName().equals("TIMEX3")) {
Timex timex = new Timex(child);
if (child.getChildCount() != 1) {
throw new RuntimeException("TIMEX3 should only contain text " + child);
}
String timexText = child.getValue();
CoreMap timexMap = new ArrayCoreMap();
// (timex)
timexMap.set(TimexAnnotation.class, timex);
// (text)
timexMap.set(CoreAnnotations.TextAnnotation.class, timexText);
// (characters)
int charBegin = offset;
timexMap.set(CoreAnnotations.CharacterOffsetBeginAnnotation.class, charBegin);
offset += timexText.length();
int charEnd = offset;
timexMap.set(CoreAnnotations.CharacterOffsetEndAnnotation.class, charEnd);
// (tokens)
if (haveTokenOffsets) {
Integer tokBegin = beginMap.get(charBegin);
int searchStep = 1; // if no exact match, search around the character offset
while (tokBegin == null) {
tokBegin = beginMap.get(charBegin - searchStep);
if (tokBegin == null) {
tokBegin = beginMap.get(charBegin + searchStep);
}
searchStep += 1;
}
searchStep = 1;
Integer tokEnd = endMap.get(charEnd);
while (tokEnd == null) {
tokEnd = endMap.get(charEnd - searchStep);
if (tokEnd == null) {
tokEnd = endMap.get(charEnd + searchStep);
}
searchStep += 1;
}
timexMap.set(CoreAnnotations.TokenBeginAnnotation.class, tokBegin);
timexMap.set(CoreAnnotations.TokenEndAnnotation.class, tokEnd);
}
// (add)
timexMaps.add(timexMap);
} else {
throw new RuntimeException("unexpected element " + child);
}
} else {
throw new RuntimeException("unexpected content " + content);
}
}
return timexMaps;
}
/**
* Label entities in an ExtractionSentence. Assumes the classifier has already been trained.
*
* @param sentence ExtractionSentence that we want to extract entities from
* @return an ExtractionSentence with text content, tree and entities set. Relations will not be
* set.
*/
private CoreMap extractEntities(CoreMap sentence, int sentCount) {
// don't add answer annotations
List<CoreLabel> testSentence =
AnnotationUtils.sentenceEntityMentionsToCoreLabels(
sentence, false, annotationsToSkip, null, useSubTypes, useBIO);
// now label the sentence
List<CoreLabel> annotatedSentence = this.classifier.classify(testSentence);
if (logger.isLoggable(Level.FINEST)) {
logger.finest("CLASSFIER OUTPUT: " + annotatedSentence);
}
List<EntityMention> extractedEntities = new ArrayList<>();
int i = 0;
// variables which keep track of partially seen entities (i.e. we've seen
// some but not all the words in them so far)
String lastType = null;
int startIndex = -1;
//
// note that labels may be in the BIO or just the IO format. we must handle both transparently
//
for (CoreLabel label : annotatedSentence) {
String type = label.get(AnswerAnnotation.class);
if (type.equals(SeqClassifierFlags.DEFAULT_BACKGROUND_SYMBOL)) {
type = null;
}
// this is an entity end boundary followed by O
if (type == null && lastType != null) {
makeEntityMention(sentence, startIndex, i, lastType, extractedEntities, sentCount);
logger.info("Found entity: " + extractedEntities.get(extractedEntities.size() - 1));
startIndex = -1;
}
// entity start preceded by an O
else if (lastType == null && type != null) {
startIndex = i;
}
// entity end followed by another entity of different type
else if (lastType != null
&& type != null
&& (type.startsWith("B-")
|| (lastType.startsWith("I-") && type.startsWith("I-") && !lastType.equals(type))
|| (notBIO(lastType) && notBIO(type) && !lastType.equals(type)))) {
makeEntityMention(sentence, startIndex, i, lastType, extractedEntities, sentCount);
logger.info("Found entity: " + extractedEntities.get(extractedEntities.size() - 1));
startIndex = i;
}
lastType = type;
i++;
}
// replace the original annotation with the predicted entities
sentence.set(MachineReadingAnnotations.EntityMentionsAnnotation.class, extractedEntities);
logger.finest("EXTRACTED ENTITIES: ");
for (EntityMention e : extractedEntities) {
if (logger.isLoggable(Level.FINEST)) {
logger.finest("\t" + e);
}
}
postprocessSentence(sentence, sentCount);
return sentence;
}