private IntPair GetNPSpanOld(IndexedWord headword, SemanticGraph dep, IList <CoreLabel> sent)
{
IndexedWord cop = dep.GetChildWithReln(headword, UniversalEnglishGrammaticalRelations.Copula);
Pair <IndexedWord, IndexedWord> leftRight = SemanticGraphUtils.LeftRightMostChildVertices(headword, dep);
// headword can be first or last word
int beginIdx = Math.Min(headword.Index() - 1, leftRight.first.Index() - 1);
int endIdx = Math.Max(headword.Index() - 1, leftRight.second.Index() - 1);
// no copula relation
if (cop == null)
{
return(new IntPair(beginIdx, endIdx));
}
// if we have copula relation
IList <IndexedWord> children = dep.GetChildList(headword);
int copIdx = children.IndexOf(cop);
if (copIdx + 1 < children.Count)
{
beginIdx = Math.Min(headword.Index() - 1, SemanticGraphUtils.LeftMostChildVertice(children[copIdx + 1], dep).Index() - 1);
}
else
{
beginIdx = headword.Index() - 1;
}
return(new IntPair(beginIdx, endIdx));
}
/// <summary>TODO: figure out how to specify where in the sentence this node goes.</summary>
/// <remarks>
/// TODO: figure out how to specify where in the sentence this node goes.
/// TODO: determine if we should be copying an IndexedWord, or working just with a FeatureLabel.
/// TODO: bombproof if this gov, dep, and reln already exist.
/// </remarks>
public override void Evaluate(SemanticGraph sg, SemgrexMatcher sm)
{
IndexedWord govNode = sm.GetNode(govNodeName);
IndexedWord newNode = new IndexedWord(newNodePrototype);
int newIndex = SemanticGraphUtils.LeftMostChildVertice(govNode, sg).Index();
// cheap En-specific hack for placing copula (beginning of governing phrase)
newNode.SetDocID(govNode.DocID());
newNode.SetIndex(newIndex);
newNode.SetSentIndex(govNode.SentIndex());
sg.AddVertex(newNode);
sg.AddEdge(govNode, newNode, relation, weight, false);
}
private void ExtractMentionForHeadword(IndexedWord headword, SemanticGraph dep, ICoreMap s, IList <Mention> mentions, ICollection <IntPair> mentionSpanSet, ICollection <IntPair> namedEntitySpanSet)
{
IList <CoreLabel> sent = s.Get(typeof(CoreAnnotations.TokensAnnotation));
SemanticGraph basic = s.Get(typeof(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation));
SemanticGraph enhanced = s.Get(typeof(SemanticGraphCoreAnnotations.EnhancedDependenciesAnnotation));
if (enhanced == null)
{
enhanced = s.Get(typeof(SemanticGraphCoreAnnotations.BasicDependenciesAnnotation));
}
// pronoun
if (headword.Tag().StartsWith("PRP"))
{
ExtractPronounForHeadword(headword, dep, s, mentions, mentionSpanSet, namedEntitySpanSet);
return;
}
// add NP mention
IntPair npSpan = GetNPSpan(headword, dep, sent);
int beginIdx = npSpan.Get(0);
int endIdx = npSpan.Get(1) + 1;
if (",".Equals(sent[endIdx - 1].Word()))
{
endIdx--;
}
// try not to have span that ends with ,
if ("IN".Equals(sent[beginIdx].Tag()))
{
beginIdx++;
}
// try to remove first IN.
AddMention(beginIdx, endIdx, headword, mentions, mentionSpanSet, namedEntitySpanSet, sent, basic, enhanced);
//
// extract the first element in conjunction (A and B -> extract A here "A and B", "B" will be extracted above)
//
// to make sure we find the first conjunction
ICollection <IndexedWord> conjChildren = dep.GetChildrenWithReln(headword, UniversalEnglishGrammaticalRelations.Conjunct);
if (conjChildren.Count > 0)
{
IndexedWord conjChild = dep.GetChildWithReln(headword, UniversalEnglishGrammaticalRelations.Conjunct);
foreach (IndexedWord c in conjChildren)
{
if (c.Index() < conjChild.Index())
{
conjChild = c;
}
}
IndexedWord left = SemanticGraphUtils.LeftMostChildVertice(conjChild, dep);
for (int endIdxFirstElement = left.Index() - 1; endIdxFirstElement > beginIdx; endIdxFirstElement--)
{
if (!sent[endIdxFirstElement - 1].Tag().Matches("CC|,"))
{
if (headword.Index() - 1 < endIdxFirstElement)
{
AddMention(beginIdx, endIdxFirstElement, headword, mentions, mentionSpanSet, namedEntitySpanSet, sent, basic, enhanced);
}
break;
}
}
}
}
