public static List<Token> groupTokens(
String text, List<Token> toks, List<? extends Span> spans, String additionalContext) {
for (int i = spans.size() - 1; i >= 0; i--) {
Span span = spans.get(i);
if (span.length() > 0) {
int s = toks.get(span.getStart()).getSpan().getStart();
int e = toks.get(span.getEnd() - 1).getSpan().getEnd();
StringBuilder lexeme = new StringBuilder();
for (int j = span.getStart(); j < span.getEnd() - 1; j++) {
lexeme.append(toks.get(j).getLexeme()).append("_");
}
lexeme.append(toks.get(span.getEnd() - 1).getLexeme());
for (int j = span.getEnd() - 1; j >= span.getStart(); j--) {
toks.remove(j);
}
Token t = new TokenCogroo(lexeme.toString(), new Span(s, e));
if (span instanceof TypedSpan) {
t.setOriginalPOSTag(((TypedSpan) span).getType());
}
if (additionalContext != null) {
t.setAdditionalContext(additionalContext);
}
toks.add(span.getStart(), t);
}
}
return toks;
}
/**
* Extracts location names from unstructured text using the named entity recognizer (NER) feature
* provided by the Apache OpenNLP Name Finder.
*
* @param plainText Contents of text document
* @return List of location names and positions
*/
public List<LocationOccurrence> extractLocationNames(String plainText) {
if (plainText == null) {
throw new IllegalArgumentException(
"plaintext input to extractLocationNames should not be null");
}
List<LocationOccurrence> nerResults = new ArrayList<LocationOccurrence>();
// The values used in these Spans are string character offsets
Span sentenceSpans[] = sentenceDetector.sentPosDetect(plainText);
// Each sentence gets processed on its own
for (Span sentenceSpan : sentenceSpans) {
// find the start and end position of this sentence in the document
String sentence = plainText.substring(sentenceSpan.getStart(), sentenceSpan.getEnd());
// tokenize the text into the required OpenNLP format
String[] tokens = tokenizer.tokenize(sentence);
// the values used in these Spans are string character offsets of each token from the sentence
// beginning
Span[] tokenPositionsWithinSentence = tokenizer.tokenizePos(sentence);
// find the location names in the tokenized text
// the values used in these Spans are NOT string character offsets, they are indices into the
// 'tokens' array
Span names[] = nameFinder.find(tokens);
// for each name that got found, create our corresponding occurrence
for (Span name : names) {
// find offsets relative to the start of the sentence
int beginningOfFirstWord = tokenPositionsWithinSentence[name.getStart()].getStart();
int endOfLastWord =
tokenPositionsWithinSentence[name.getEnd() - 1]
.getEnd(); // -1 because the high end of a Span is noninclusiv
// to get offsets relative to the document as a whole, just add the offset for the sentence
// itself
int startOffsetInDoc = sentenceSpan.getStart() + beginningOfFirstWord;
int endOffsetInDoc = sentenceSpan.getStart() + endOfLastWord;
// look back into the original input string to figure out what the text is that I got a hit
// on
String nameInDocument = plainText.substring(startOffsetInDoc, endOffsetInDoc);
// add to List of results to return
nerResults.add(new LocationOccurrence(nameInDocument, startOffsetInDoc));
}
}
// this is necessary to maintain consistent results across
// multiple runs on the same data, which is what we want
nameFinder.clearAdaptiveData();
return nerResults;
}
public static List<Token> groupTokensChar(
String text, List<Token> toks, List<Span> charSpans, String additionalContext) {
if (charSpans == null || charSpans.size() == 0) {
return toks;
}
int lastVisitedTok = 0;
List<Span> spans = new ArrayList<Span>(charSpans.size());
for (Span ch : charSpans) {
// System.out.println("looking for: " + ch.getCoveredText(text));
Token aToken = toks.get(lastVisitedTok);
while (aToken.getSpan().getStart() < ch.getStart()) {
lastVisitedTok++;
aToken = toks.get(lastVisitedTok);
}
int start = lastVisitedTok;
while (aToken.getSpan().getEnd() < ch.getEnd()) {
lastVisitedTok++;
aToken = toks.get(lastVisitedTok);
}
int end = lastVisitedTok + 1;
Span tokSpan = new Span(start, end);
spans.add(tokSpan);
}
return groupTokens(text, toks, spans, additionalContext);
}
public void execute() throws ExecutionException {
AnnotationSet outputAS = document.getAnnotations(annotationSetName);
String text = document.getContent().toString();
Span[] tokens = tokenizer.getTokens(text);
try {
for (Span token : tokens) {
FeatureMap features = Factory.newFeatureMap();
features.put(
ANNIEConstants.TOKEN_STRING_FEATURE_NAME,
text.substring(token.getStart(), token.getEnd()));
outputAS.add(
(long) token.getStart(),
(long) token.getEnd(),
ANNIEConstants.TOKEN_ANNOTATION_TYPE,
features);
}
} catch (Exception e) {
throw new ExecutionException("error running tokenizer", e);
}
}
private List<Section> detectSentences(AnalysedText at, String language) {
SentenceDetector sentenceDetector = getSentenceDetector(language);
List<Section> sentences;
if (sentenceDetector != null) {
sentences = new ArrayList<Section>();
for (opennlp.tools.util.Span sentSpan : sentenceDetector.sentPosDetect(at.getSpan())) {
Sentence sentence = at.addSentence(sentSpan.getStart(), sentSpan.getEnd());
log.trace(" > add {}", sentence);
sentences.add(sentence);
}
} else {
sentences = null;
}
return sentences;
}
/**
* Adds named entity information to parses.
*
* @param tag named entity type
* @param names spans of tokens that are named entities
* @param tokens parses for the tokens
*/
private static void addNames(String tag, List names, Parse[] tokens) {
for (int i = 0; i < names.size(); i++) {
Span nameTokenSpan = (Span) names.get(i);
Parse startToken = tokens[nameTokenSpan.getStart()];
Parse endToken = tokens[nameTokenSpan.getEnd()];
Parse commonP = startToken.getCommonParent(endToken);
if (commonP != null) {
Span nameSpan = new Span(startToken.getSpan().getStart(), endToken.getSpan().getEnd());
if (nameSpan.equals(commonP.getSpan())) {
// common parent matches exactly the named entity
commonP.insert(new Parse(commonP.getText(), nameSpan, tag, 1.0));
} else {
// common parent includes the named entity
Parse[] kids = commonP.getChildren();
boolean crossingKids = false;
for (int j = 0; j < kids.length; j++)
if (nameSpan.crosses(kids[j].getSpan())) crossingKids = true;
if (!crossingKids) {
// named entity does not cross children
commonP.insert(new Parse(commonP.getText(), nameSpan, tag, 1.0));
} else {
// NE crosses children
if (commonP.getType().equals("NP")) {
Parse[] grandKids = kids[0].getChildren();
Parse last = grandKids[grandKids.length - 1];
if (grandKids.length > 1 && nameSpan.contains(last.getSpan()))
commonP.insert(new Parse(commonP.getText(), commonP.getSpan(), tag, 1.0));
}
}
}
}
}
}