public StringList read() throws IOException {
String line = lineStream.read();
StringList name = null;
if ((line != null) && (!StringUtil.isEmpty(line))) {
String name2;
// find the location of the name separator in the line of data.
int pos = line.indexOf(' ');
if ((pos != -1)) {
String parsed = line.substring(0, pos);
// the data is in ALL CAPS ... so the easiest way is to convert
// back to standard mixed case.
if ((parsed.length() > 2) && (parsed.startsWith("MC"))) {
name2 =
parsed.substring(0, 1).toUpperCase(locale)
+ parsed.substring(1, 2).toLowerCase(locale)
+ parsed.substring(2, 3).toUpperCase(locale)
+ parsed.substring(3).toLowerCase(locale);
} else {
name2 =
parsed.substring(0, 1).toUpperCase(locale) + parsed.substring(1).toLowerCase(locale);
}
name = new StringList(new String[] {name2});
}
}
return name;
}
public Span[] tokenizePos(String s) {
boolean isWhitespace;
List<Span> tokens = new ArrayList<Span>();
int sl = s.length();
int start = -1;
char pc = 0;
for (int ci = 0; ci <= sl; ci++) {
char c = ci < sl ? s.charAt(ci) : ' ';
isWhitespace = StringUtil.isWhitespace(c);
if (!isWhitespace & start < 0) { // new token starts
start = ci;
}
if (isWhitespace && start >= 0) { // end of token
// limited support for punctations at the end of words
if (start < ci - 1
&& (pc == '.' || pc == ',' || pc == '!' || pc == '?' || pc == ';' || pc == ':')) {
tokens.add(new Span(start, ci - 1));
tokens.add(new Span(ci - 1, ci));
} else {
tokens.add(new Span(start, ci));
}
start = -1;
}
}
return tokens.toArray(new Span[tokens.size()]);
}
public static void populatePOSDictionary(
ObjectStream<POSSample> samples, MutableTagDictionary dict, int cutoff) throws IOException {
System.out.println("Expanding POS Dictionary ...");
long start = System.nanoTime();
// the data structure will store the word, the tag, and the number of
// occurrences
Map<String, Map<String, AtomicInteger>> newEntries =
new HashMap<String, Map<String, AtomicInteger>>();
POSSample sample;
while ((sample = samples.read()) != null) {
String[] words = sample.getSentence();
String[] tags = sample.getTags();
for (int i = 0; i < words.length; i++) {
// only store words
if (!StringPattern.recognize(words[i]).containsDigit()) {
String word;
if (dict.isCaseSensitive()) {
word = words[i];
} else {
word = StringUtil.toLowerCase(words[i]);
}
if (!newEntries.containsKey(word)) {
newEntries.put(word, new HashMap<String, AtomicInteger>());
}
String[] dictTags = dict.getTags(word);
if (dictTags != null) {
for (String tag : dictTags) {
// for this tags we start with the cutoff
Map<String, AtomicInteger> value = newEntries.get(word);
if (!value.containsKey(tag)) {
value.put(tag, new AtomicInteger(cutoff));
}
}
}
if (!newEntries.get(word).containsKey(tags[i])) {
newEntries.get(word).put(tags[i], new AtomicInteger(1));
} else {
newEntries.get(word).get(tags[i]).incrementAndGet();
}
}
}
}
// now we check if the word + tag pairs have enough occurrences, if yes we
// add it to the dictionary
for (Entry<String, Map<String, AtomicInteger>> wordEntry : newEntries.entrySet()) {
List<String> tagsForWord = new ArrayList<String>();
for (Entry<String, AtomicInteger> entry : wordEntry.getValue().entrySet()) {
if (entry.getValue().get() >= cutoff) {
tagsForWord.add(entry.getKey());
}
}
if (tagsForWord.size() > 0) {
dict.put(wordEntry.getKey(), tagsForWord.toArray(new String[tagsForWord.size()]));
}
}
System.out.println(
"... finished expanding POS Dictionary. [" + (System.nanoTime() - start) / 1000000 + "ms]");
}
public NameSample read() throws IOException {
List<String> tokens = new ArrayList<String>();
List<String> neTypes = new ArrayList<String>();
boolean isClearAdaptiveData = false;
// Empty line indicates end of sentence
String line;
while ((line = lineStream.read()) != null && !StringUtil.isEmpty(line)) {
// clear adaptive data if document mark appears following
// CoNLL03 conventions
if (clearFeatures.equalsIgnoreCase("docstart") && line.startsWith("-DOCSTART-")) {
isClearAdaptiveData = true;
String emptyLine = lineStream.read();
if (!StringUtil.isEmpty(emptyLine))
throw new IOException("Empty line after -DOCSTART- not empty: '" + emptyLine + "'!");
continue;
}
String fields[] = line.split("\t");
if (fields.length == 2) {
tokens.add(fields[0]);
neTypes.add(fields[1]);
} else {
throw new IOException(
"Expected two fields per line in training data, got "
+ fields.length
+ " for line '"
+ line
+ "'!");
}
}
// if no -DOCSTART- mark, check if we need to clear features every sentence
if (clearFeatures.equalsIgnoreCase("yes")) {
isClearAdaptiveData = true;
}
if (tokens.size() > 0) {
// convert name tags into spans
List<Span> names = new ArrayList<Span>();
int beginIndex = -1;
int endIndex = -1;
for (int i = 0; i < neTypes.size(); i++) {
String neTag = neTypes.get(i);
if (neTag.equals("O")) {
// O means we don't have anything this round.
if (beginIndex != -1) {
names.add(extract(beginIndex, endIndex, neTypes.get(beginIndex)));
beginIndex = -1;
endIndex = -1;
}
} else if (neTag.startsWith("B-")) {
// B- prefix means we have two same entities of the same class next to each other
if (beginIndex != -1) {
names.add(extract(beginIndex, endIndex, neTypes.get(beginIndex)));
}
beginIndex = i;
endIndex = i + 1;
} else if (neTag.startsWith("I-")) {
// I- starts or continues a current name entity
if (beginIndex == -1) {
beginIndex = i;
endIndex = i + 1;
} else if (!neTag.endsWith(neTypes.get(beginIndex).substring(1))) {
// we have a new tag type following a tagged word series
// also may not have the same I- starting the previous!
names.add(extract(beginIndex, endIndex, neTypes.get(beginIndex)));
beginIndex = i;
endIndex = i + 1;
} else {
endIndex++;
}
} else {
throw new IOException("Invalid tag: " + neTag);
}
}
// if one span remains, create it here
if (beginIndex != -1) names.add(extract(beginIndex, endIndex, neTypes.get(beginIndex)));
return new NameSample(
tokens.toArray(new String[tokens.size()]),
names.toArray(new Span[names.size()]),
isClearAdaptiveData);
} else if (line != null) {
// Just filter out empty events, if two lines in a row are empty
return read();
} else {
// source stream is not returning anymore lines
return null;
}
}
