@Override
public Sequence<IString> process(String input) {
String tokenizerInput = toUncased(input.trim());
Tokenizer<CoreLabel> tokenizer = tf.getTokenizer(new StringReader(tokenizerInput));
List<String> outputStrings = new LinkedList<>();
while (tokenizer.hasNext()) {
String string = tokenizer.next().get(TextAnnotation.class);
outputStrings.add(string);
}
return IStrings.toIStringSequence(outputStrings);
}
@Override
public SymmetricalWordAlignment processAndAlign(String input) {
input = input.trim();
// Run through the tokenizer and convert to sequence
String tokenizerInput = toUncased(input);
String[] uncasedInputTokens = tokenizerInput.split("\\s+");
Tokenizer<CoreLabel> tokenizer = tf.getTokenizer(new StringReader(tokenizerInput));
List<CoreLabel> outputTokens = tokenizer.tokenize();
IString[] outputSequence = new IString[outputTokens.size()];
for (int i = 0; i < outputSequence.length; ++i) {
String outputToken = outputTokens.get(i).get(TextAnnotation.class);
outputSequence[i] = new IString(outputToken);
}
// Whitespace tokenization of input, create alignment
Sequence<IString> inputSequence = IStrings.tokenize(input);
assert inputSequence.size() == uncasedInputTokens.length;
SymmetricalWordAlignment alignment =
new SymmetricalWordAlignment(
inputSequence, new SimpleSequence<IString>(true, outputSequence));
// Generate the alignments
StringBuilder inputToken = new StringBuilder();
for (int i = 0, j = 0, limit = outputTokens.size(); j < limit; ++j) {
CoreLabel tokenizedToken = outputTokens.get(j);
String inputTokenPart = toUncased(tokenizedToken.get(OriginalTextAnnotation.class));
alignment.addAlign(i, j);
inputToken.append(inputTokenPart);
if (i >= uncasedInputTokens.length) {
System.err.println("WARNING: Non-invertible input: " + input);
break;
}
if (uncasedInputTokens[i].equals(inputToken.toString())) {
++i;
inputToken = new StringBuilder();
}
}
return alignment;
}
private static List<TypedDependency> getDependencies(String sentence) {
if (pipeline == null) {
loadModels();
}
TokenizerFactory<CoreLabel> tokenizerFactory =
PTBTokenizer.factory(new CoreLabelTokenFactory(), "");
Tokenizer<CoreLabel> tok = tokenizerFactory.getTokenizer(new StringReader(sentence));
List<CoreLabel> rawWords2 = tok.tokenize();
Tree parse = lp.apply(rawWords2);
// parse.pennPrint();
//
// System.out.println(parse.toString());
TreebankLanguagePack tlp = lp.treebankLanguagePack(); // PennTreebankLanguagePack for English
GrammaticalStructureFactory gsf = tlp.grammaticalStructureFactory();
GrammaticalStructure gs = gsf.newGrammaticalStructure(parse);
List<TypedDependency> tdl = gs.typedDependenciesCCprocessed();
return tdl;
}
private Tree getPosTree(String sentence) {
final Tokenizer<CoreLabel> tokenizer =
tokenizerFactory.getTokenizer(new StringReader(sentence));
final List<CoreLabel> tokens = tokenizer.tokenize();
return parser.apply(tokens);
}
@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);
}
private static List<CoreLabel> tokenize(String str) {
Tokenizer<CoreLabel> tokenizer = tokenizerFactory.getTokenizer(new StringReader(str));
return tokenizer.tokenize();
}