@Override
public String getLexiconURL(Predicate pred) {
// dunno if this still works
return "http://clic.ub.edu/mbertran/ancora/lexentry.php?file="
+ pred.getLemma()
+ ".lex.xml&lexicon=AnCoraVerb_ES";
}
@Override
public String getCoreArgumentLabelSequence(Predicate pred, Map<Word, String> proposition) {
Sentence s = pred.getMySentence();
StringBuffer ret = new StringBuffer();
for (int i = 1; i < s.size(); ++i) {
Word w = s.get(i);
if (pred == w) {
ret.append(" ").append(pred.getSense());
if (useVoice) ret.append(isPassive(pred) ? "/P" : "/A");
}
if (proposition.containsKey(w)) {
Matcher m = CALSPattern.matcher(proposition.get(w));
if (m.matches()) ret.append(" ").append(m.group(1));
}
}
return ret.toString();
}
@Override
public String getDefaultSense(Predicate pred) {
String PFeat = pred.getFeats();
String label;
if (PFeat.contains("postype=auxiliary")) {
label = "c2";
} else if (PFeat.contains("postype=common")) {
label = "a2";
} else if (PFeat.contains("postype=main")) {
label = "a2";
} else if (PFeat.contains("postype=qualificative")) {
label = "b2";
} else if (PFeat.contains("postype=semiauxiliary")) {
label = "c2";
} else {
label = "a2";
}
return pred.getLemma() + "." + label;
}
public void write(Sentence s) {
try {
for (Predicate p : s.getPredicates()) {
if (p.getSense().equals("Action") || p.getSense().equals("OPERATION")) {
out.write(
id(p)
+ "\t"
+ "Action"
+ " "
+ p.getBegin()
+ " "
+ p.getEnd()
+ "\t"
+ p.getForm()
+ "\n");
for (Word w : p.getArgMap().keySet()) {
String label = p.getArgMap().get(w);
if (label.equals("Theme")) label = "Object";
if (!word2id.containsKey(w))
out.write(
id(w)
+ "\t"
+ label
+ " "
+ w.getBegin()
+ " "
+ w.getEnd()
+ "\t"
+ w.getForm()
+ "\n");
out.write(
"R"
+ (rnum++)
+ "\t"
+ (label.equals("Actor")
? ("IsActorOf Arg1:" + id(w) + " Arg2:" + id(p))
: (label.equals("Property")
? ("HasProperty Arg1:" + id(p) + " Arg2:" + id(w))
: ("ActsOn Arg1:" + id(p) + " Arg2:" + id(w))))
+ "\n");
}
}
if (p.getSense().equals("Object")
|| p.getSense().equals("CONCEPT")
|| p.getSense().equals("Property")) {
if (!word2id.containsKey(p))
out.write(
id(p)
+ "\t"
+ p.getSense()
+ " "
+ p.getBegin()
+ " "
+ p.getEnd()
+ "\t"
+ p.getForm()
+ "\n");
for (Word w : p.getArgMap().keySet()) {
String label = p.getArgMap().get(w);
if (label.equals("Theme")) label = "Object";
if (!word2id.containsKey(w))
out.write(
id(w)
+ "\t"
+ label
+ " "
+ w.getBegin()
+ " "
+ w.getEnd()
+ "\t"
+ w.getForm()
+ "\n");
out.write(
"R" + (rnum++) + "\t" + "HasProperty Arg1:" + id(p) + " Arg2:" + id(w) + "\n");
}
}
}
// out.write(s.toString()+"\n\n");
} catch (Exception e) {
e.printStackTrace();
System.out.println("Failed to write sentance.");
System.exit(1);
}
}
public void write(Sentence s) {
StringBuffer text = new StringBuffer();
/** hard-coded fix for test set starting with "Simply" * */
if (s.get(1).getForm().equals("Simply")) {
// System.err.println("XXX");
text.append(" ");
}
for (int i = 1; i < s.size(); i++) {
if (i > 1) text.append(' ');
s.get(i).setBegin(text.length());
text.append(s.get(i).getForm());
s.get(i).setEnd(text.length());
}
int annID = count * 100;
try {
out.write(" <sentence ID=\"" + count + "\">\n");
out.write(" <text>" + text + "</text>\n");
out.write(" <annotationSets>\n");
for (int i = 1; i < s.size(); i++) {
if (s.get(i) instanceof Predicate) {
Predicate p = (Predicate) s.get(i);
int layerID = annID * 100 + 1;
int labelID = layerID * 100 + 1;
out.write(
" <annotationSet ID=\""
+ annID
+ "\" frameName=\""
+ p.getSense()
+ "\">\n");
out.write(" <layers>\n");
out.write(" <layer ID=\"" + layerID + "\" name=\"Target\">\n");
out.write(" <labels>\n");
out.write(
" <label ID=\""
+ labelID
+ "\" end=\""
+ (p.getEnd() - 1)
+ "\" name=\"Target\" start=\""
+ p.getBegin()
+ "\"/>\n");
out.write(" </labels>\n");
out.write(" </layer>\n");
layerID++;
labelID = layerID * 100 + 1;
out.write(" <layer ID=\"" + layerID + "\" name=\"FE\">\n");
out.write(" <labels>\n");
if (!reconstructSpan)
for (Word a : p.getArgMap().keySet()) {
out.write(
" <label ID=\""
+ labelID
+ "\" end=\""
+ (a.getEnd() - 1)
+ "\" name=\""
+ p.getArgMap().get(a)
+ "\" start=\""
+ a.getBegin()
+ "\"/>\n");
labelID++;
}
else {
Map<String, List<Word[]>> label2spans = new HashMap<String, List<Word[]>>();
for (Word a : p.getArgMap().keySet()) {
Word[] begin_end = DasFilter.pass(p, a);
if (begin_end == null) continue;
String label = p.getArgMap().get(a);
if (!label2spans.containsKey(label)) label2spans.put(label, new LinkedList<Word[]>());
label2spans.get(label).add(begin_end);
}
Map<String, List<Word[]>> newlabel2spans = new HashMap<String, List<Word[]>>();
for (String label : label2spans.keySet()) {
List<Word[]> begin_ends = DasFilter.merge(label2spans.get(label));
newlabel2spans.put(label, begin_ends);
}
DasFilter.resolveConflictingSpans(newlabel2spans);
for (String label : newlabel2spans.keySet()) {
List<Word[]> begin_ends = newlabel2spans.get(label);
for (Word[] begin_end : begin_ends) {
out.write(
" <label ID=\""
+ labelID
+ "\" end=\""
+ (begin_end[1].getEnd() - 1)
+ "\" name=\""
+ label
+ "\" start=\""
+ (begin_end[0].getBegin())
+ "\"/>\n");
labelID++;
}
}
}
out.write(" </labels>\n");
out.write(" </layer>\n");
out.write(" </layers>\n");
out.write(" </annotationSet>\n");
}
}
out.write(" </annotationSets>\n");
out.write(" </sentence>\n");
count++;
} catch (IOException e) {
e.printStackTrace();
System.out.println("Failed to write sentance.");
System.exit(1);
}
}
/**
* This is the code we used to deduce voice in Spanish (and Catalan) for CoNLL 2009, however we
* didn't actually use it in the final submission. I think it was because we never saw any real
* improvement. I'm not sure it's proper though, my Spanish skills are rather non-existant. I just
* put it here for future reference.
*
* @param pred the predicate
* @return true if the predicate (verb) is in passive tense, false otherwise
*/
private boolean isPassive(Predicate pred) {
for (Word c : pred.getChildren())
if ((c.getLemma().equals("estar") || c.getLemma().equals("ser"))
&& c.getFeats().contains("auxiliary")) return true;
return false;
}