/**
* Method that computes all the features for each word in a sentence. For more information, read
* Espla-Gomis, M., Sánchez-Martínez, F. Forcada, M.L. 2011. In Proceedings of the XIII Machine
* Translation Summit, Xiamen University, Xiamen, China.
*
* @param sourcesegment Segment in source language to be translated
* @param tu Translation unit from which features should be computed
* @param segmaxlen Maximum size of the segments to be created
* @param score Score obtained for the matching between sourcesentence and matchsentence
* @param debug Flag that indicates if debug messages should be shown or not
* @return Returns all the features for each word in a sentence as a list of pairs word-features
* array
*/
public List<Pair<String, double[]>> Compute(
PrintWriter debugpw, boolean discard_words_without_evidence) {
List<Pair<String, double[]>> features = new LinkedList<Pair<String, double[]>>();
// Set<Segment> sourcesegmented=sourcesentence.AllSegmentsInSentence(segmaxlen);
boolean[] matching = new boolean[tu.getSource().getSentence().size()];
Segment.EditDistance(
sourcesegment.getSentenceCodes(), tu.getSource().getSentenceCodes(), matching, false);
for (int wpos = 0; wpos < tu.getTarget().getSentence().size(); wpos++) {
String word = tu.getTarget().getWord(wpos).getValue();
double[] feat_tmp;
if (discard_words_without_evidence) {
feat_tmp = new double[segmaxlen * segmaxlen * 2];
} else {
feat_tmp = new double[segmaxlen * segmaxlen * 2 + 1];
}
int covering_evidence = 0;
if (debugpw != null) {
debugpw.print(word);
debugpw.print(": ");
}
Set<Evidence> evidenceset = new LinkedHashSet<Evidence>();
for (int mlen = 0; mlen < segmaxlen; mlen++) {
for (int nlen = 0; nlen < segmaxlen; nlen++) {
int coveringintersec_p = 0;
// Set<Evidence> coveringintersec_p=new LinkedHashSet<Evidence>();
int coveringintersec_n = 0;
// Set<Evidence> coveringintersec_n=new LinkedHashSet<Evidence>();
for (Evidence evidence : tu.getEvidences()) {
// If the target side covers the word an the source side
// is of the right lenght
if (Evidence.Cover(
wpos,
evidence.getTranslation().getPosition(),
evidence.getTranslation().getLength())) {
if (evidence.getSegment().getLength() == (mlen + 1)
&& evidence.getTranslation().getLength() == (nlen + 1)) {
if (debugpw != null) {
evidenceset.add(evidence);
}
for (int i = 0; i < evidence.getSegment().getLength(); i++) {
if (matching[evidence.getSegment().getPosition() + i]) {
coveringintersec_p++;
} else {
coveringintersec_n++;
}
}
}
}
}
if (coveringintersec_p > 0 || coveringintersec_n > 0) {
covering_evidence++;
}
// Positive source
feat_tmp[mlen * segmaxlen * 2 + nlen * 2] = (double) coveringintersec_p;
// Negative source
feat_tmp[mlen * segmaxlen * 2 + nlen * 2 + 1] = (double) coveringintersec_n;
}
}
if (debugpw != null) {
for (Evidence e : evidenceset) {
debugpw.print(e.toString());
debugpw.print(" ");
}
debugpw.println();
}
if (!discard_words_without_evidence) {
feat_tmp[feat_tmp.length - 1] = score;
features.add(new Pair(word, feat_tmp));
} else {
if (covering_evidence > 0) {
features.add(new Pair(word, feat_tmp));
} else {
features.add(new Pair(word, null));
}
}
}
return features;
}
