@Override
public void process(JCas jcas) throws AnalysisEngineProcessException {
UIMAProfiler.getProfiler("AnalysisEngine").start(this, "process");
Lemmatizer mateLemmatizer = mateLemmatizerModel.getEngine();
Tagger mateTagger = mateTaggerModel.getEngine();
/*
* keeps an array of annotations in memory so as to be able
* to access them by index.
*/
List<WordAnnotation> annotations = Lists.newArrayList();
FSIterator<Annotation> it = jcas.getAnnotationIndex(WordAnnotation.type).iterator();
while (it.hasNext()) {
WordAnnotation a = (WordAnnotation) it.next();
annotations.add(a);
}
String[] tokens = new String[annotations.size() + 2];
// preprends to fake words to prevent Mate from bugging on the two first words
tokens[0] = "<root>";
tokens[1] = "<root2>";
for (int i = 0; i < annotations.size(); i++)
tokens[i + 2] = annotations.get(i).getCoveredText();
SentenceData09 mateSentence = new SentenceData09();
mateSentence.init(tokens);
// Run POS tagging
mateSentence = mateTagger.apply(mateSentence);
// Run lemmatization
mateSentence = mateLemmatizer.apply(mateSentence);
WordAnnotation wordAnnotation;
for (int j = 1; j < mateSentence.length(); j++) {
wordAnnotation = annotations.get(j - 1);
wordAnnotation.setTag(mateSentence.ppos[j]);
wordAnnotation.setLemma(mateSentence.plemmas[j]);
}
UIMAProfiler.getProfiler("AnalysisEngine").stop(this, "process");
}
/**
* Create an instance without root of the input instance
*
* @param instance
*/
public SentenceData09(SentenceData09 i) {
int length = i.length() - 1;
forms = new String[length];
gpos = new String[length];
ppos = new String[length];
plemmas = new String[length];
plabels = new String[length];
lemmas = new String[length];
heads = new int[length];
pheads = new int[length];
ofeats = new String[length];
pfeats = new String[length];
labels = new String[length];
fillp = new String[length];
id = new String[length];
for (int j = 0; j < length; j++) {
forms[j] = i.forms[j + 1];
ppos[j] = i.ppos[j + 1];
gpos[j] = i.gpos[j + 1];
labels[j] = i.labels[j + 1];
heads[j] = i.heads[j + 1];
if (i.pheads != null) pheads[j] = i.pheads[j + 1];
if (i.plabels != null) plabels[j] = i.plabels[j + 1];
if (i.lemmas != null) lemmas[j] = i.lemmas[j + 1];
plemmas[j] = i.plemmas[j + 1];
if (i.ofeats != null) ofeats[j] = i.ofeats[j + 1];
if (i.pfeats != null) pfeats[j] = i.pfeats[j + 1];
if (i.fillp != null) fillp[j] = i.fillp[j + 1];
if (i.id != null) id[j] = i.id[j + 1];
}
}
public static Results evaluate(String act_file, String pred_file) throws Exception {
CONLLReader09 goldReader = new CONLLReader09(act_file, -1);
CONLLReader09 predictedReader = new CONLLReader09(pred_file, -1);
int total = 0, corr = 0, corrL = 0;
int numsent = 0, corrsent = 0, corrsentL = 0;
SentenceData09 goldInstance = goldReader.getNext();
SentenceData09 predInstance = predictedReader.getNext();
while (goldInstance != null) {
int instanceLength = goldInstance.length();
if (instanceLength != predInstance.length()) {
Parser.out.println("Lengths do not match on sentence " + numsent);
}
int[] goldHeads = goldInstance.heads;
String[] goldLabels = goldInstance.labels;
int[] predHeads = predInstance.pheads;
String[] predLabels = predInstance.plabels;
boolean whole = true;
boolean wholeL = true;
// NOTE: the first item is the root info added during nextInstance(), so we skip it.
int punc = 0;
for (int i = 1; i < instanceLength; i++) {
if (predHeads[i] == goldHeads[i]) {
corr++;
if (goldLabels[i].equals(predLabels[i])) {
corrL++;
} else {
// Parser.out.println(numsent+" error gold "+goldLabels[i]+" "+predLabels[i]+" head
// "+goldHeads[i]+" child "+i);
wholeL = false;
}
} else {
// Parser.out.println(numsent+"error gold "+goldLabels[i]+" "+predLabels[i]+" head
// "+goldHeads[i]+" child "+i);
whole = false;
wholeL = false;
}
}
total += ((instanceLength - 1) - punc); // Subtract one to not score fake root token
if (whole) {
corrsent++;
}
if (wholeL) {
corrsentL++;
}
numsent++;
goldInstance = goldReader.getNext();
predInstance = predictedReader.getNext();
}
Results r = new Results();
r.total = total;
r.corr = corr;
r.las = (float) Math.round(((double) corrL / total) * 100000) / 1000;
r.ula = (float) Math.round(((double) corr / total) * 100000) / 1000;
Parser.out.print("Total: " + total + " \tCorrect: " + corr + " ");
Parser.out.println(
"LAS: "
+ (double) Math.round(((double) corrL / total) * 100000) / 1000
+ " \tTotal: "
+ (double) Math.round(((double) corrsentL / numsent) * 100000) / 1000
+ " \tULA: "
+ (double) Math.round(((double) corr / total) * 100000) / 1000
+ " \tTotal: "
+ (double) Math.round(((double) corrsent / numsent) * 100000) / 1000);
return r;
}
public void createWithRoot(SentenceData09 i) {
int length = i.length();
int offset = 0;
if (!i.forms[0].equals(CONLLReader09.ROOT)) {
length++;
offset = -1;
}
forms = new String[length];
gpos = new String[length];
ppos = new String[length];
plemmas = new String[length];
plabels = new String[length];
lemmas = new String[length];
heads = new int[length];
pheads = new int[length];
ofeats = new String[length];
pfeats = new String[length];
labels = new String[length];
fillp = new String[length];
id = new String[length];
feats = new String[forms.length][];
for (int j = 1; j < length; j++) {
forms[j] = i.forms[j + offset];
ppos[j] = i.ppos[j + offset];
gpos[j] = i.gpos[j + offset];
labels[j] = i.labels[j + offset];
heads[j] = i.heads[j + offset];
if (i.pheads != null) pheads[j] = i.pheads[j + offset];
if (i.plabels != null) plabels[j] = i.plabels[j + offset];
if (i.lemmas != null) lemmas[j] = i.lemmas[j + offset];
plemmas[j] = i.plemmas[j + offset];
// if (i.ofeats!=null) ofeats[j] = i.ofeats[j+offset];
ofeats[j] = i.ofeats[j + offset].equals(CONLLWriter09.DASH) ? "_" : i.ofeats[j + offset];
// if (i.pfeats!=null) pfeats[j] = i.pfeats[j+offset];
if (i.pfeats != null && i.pfeats[j + offset] != null) {
if (i.pfeats[j + offset].equals(CONLLWriter09.DASH)) feats[j] = null;
else {
feats[j] = i.pfeats[j + offset].split(CONLLReader09.PIPE);
// if (info[7].equals(CONLLWriter09.DASH)) it.feats[i]=null;
// else {
// it.feats[i] =info[7].split(PIPE);
pfeats[j] = i.pfeats[j + offset];
// }
}
}
if (i.fillp != null) fillp[j] = i.fillp[j + offset];
if (i.id != null) id[j] = i.id[j + offset];
}
forms[0] = CONLLReader09.ROOT;
plemmas[0] = CONLLReader09.ROOT_LEMMA;
fillp[0] = "N";
lemmas[0] = CONLLReader09.ROOT_LEMMA;
gpos[0] = CONLLReader09.ROOT_POS;
ppos[0] = CONLLReader09.ROOT_POS;
labels[0] = CONLLReader09.NO_TYPE;
heads[0] = -1;
plabels[0] = CONLLReader09.NO_TYPE;
pheads[0] = -1;
ofeats[0] = CONLLReader09.NO_TYPE;
id[0] = "0";
}