// TODO: never used?
public virtual void ConstructSentenceRelationMentions()
{
//
// construct the mRelationEntityMentions matrix
//
ICollection <string> relKeys = mRelationMentions.Keys;
foreach (string key in relKeys)
{
AceRelationMention rm = mRelationMentions[key];
int sentence = mTokens[rm.GetArg(0).GetHead().GetTokenStart()].GetSentence();
//
// no need to adjust the number of rows: was done in parseDocument
//
// store the relation mentions in increasing order
// (a) of the start position of their head, or
// (b) if start is the same, in increasing order of ends
List <AceRelationMention> sentRels = mSentenceRelationMentions[sentence];
bool added = false;
for (int i = 0; i < sentRels.Count; i++)
{
AceRelationMention crt = sentRels[i];
if ((crt.GetMinTokenStart() > rm.GetMinTokenStart()) || (crt.GetMinTokenStart() == rm.GetMinTokenStart() && crt.GetMaxTokenEnd() > rm.GetMaxTokenEnd()))
{
sentRels.Add(i, rm);
added = true;
break;
}
}
if (!added)
{
sentRels.Add(rm);
}
}
}
/// <summary>Parses an ACE document.</summary>
/// <remarks>
/// Parses an ACE document. Works in the following steps: (a) reads both the
/// XML annotations; (b) reads the tokens; (c) matches the tokens against the
/// annotations (d) constructs mSentenceEntityMentions and
/// mRelationEntityMentions
/// </remarks>
/// <exception cref="System.IO.IOException"/>
/// <exception cref="Org.Xml.Sax.SAXException"/>
/// <exception cref="Javax.Xml.Parsers.ParserConfigurationException"/>
public static Edu.Stanford.Nlp.IE.Machinereading.Domains.Ace.Reader.AceDocument ParseDocument(string prefix, bool usePredictedBoundaries)
{
mLog.Fine("Reading document " + prefix);
Edu.Stanford.Nlp.IE.Machinereading.Domains.Ace.Reader.AceDocument doc = null;
//
// read the ACE XML annotations
//
if (usePredictedBoundaries == false)
{
doc = AceDomReader.ParseDocument(new File(prefix + XmlExt));
}
else
{
// log.info("Parsed " + doc.getEntityMentions().size() +
// " entities in document " + prefix);
//
// will use the predicted entity boundaries (see below)
//
int lastSlash = prefix.LastIndexOf(File.separator);
System.Diagnostics.Debug.Assert((lastSlash > 0 && lastSlash < prefix.Length - 1));
string id = Sharpen.Runtime.Substring(prefix, lastSlash + 1);
// log.info(id + ": " + prefix);
doc = new Edu.Stanford.Nlp.IE.Machinereading.Domains.Ace.Reader.AceDocument(id);
}
doc.SetPrefix(prefix);
//
// read the raw byte stream
//
string trueCasedFileName = prefix + OrigExt + ".truecase";
if ((new File(trueCasedFileName).Exists()))
{
mLog.Severe("Using truecased file: " + trueCasedFileName);
doc.ReadRawBytes(trueCasedFileName);
}
else
{
doc.ReadRawBytes(prefix + OrigExt);
}
//
// read the AceTokens
//
int offsetToSubtract = 0;
IList <IList <AceToken> > sentences = AceSentenceSegmenter.TokenizeAndSegmentSentences(prefix);
doc.SetSentences(sentences);
foreach (IList <AceToken> sentence in sentences)
{
foreach (AceToken token in sentence)
{
offsetToSubtract = token.AdjustPhrasePositions(offsetToSubtract, token.GetLiteral());
doc.AddToken(token);
}
}
//
// match char sequences to phrases
//
doc.MatchCharSeqs(prefix);
//
// construct the mEntityMentions matrix
//
ICollection <string> entityKeys = doc.mEntityMentions.Keys;
int sentence_1;
foreach (string key in entityKeys)
{
AceEntityMention em = doc.mEntityMentions[key];
sentence_1 = doc.mTokens[em.GetHead().GetTokenStart()].GetSentence();
// adjust the number of rows if necessary
while (sentence_1 >= doc.mSentenceEntityMentions.Count)
{
doc.mSentenceEntityMentions.Add(new List <AceEntityMention>());
doc.mSentenceRelationMentions.Add(new List <AceRelationMention>());
doc.mSentenceEventMentions.Add(new List <AceEventMention>());
}
// store the entity mentions in increasing order:
// (a) of the start position of their head
// (b) if start is the same, in increasing order of the head end
List <AceEntityMention> sentEnts = doc.mSentenceEntityMentions[sentence_1];
bool added = false;
for (int i = 0; i < sentEnts.Count; i++)
{
AceEntityMention crt = sentEnts[i];
if ((crt.GetHead().GetTokenStart() > em.GetHead().GetTokenStart()) || (crt.GetHead().GetTokenStart() == em.GetHead().GetTokenStart() && crt.GetHead().GetTokenEnd() > em.GetHead().GetTokenEnd()))
{
sentEnts.Add(i, em);
added = true;
break;
}
}
if (!added)
{
sentEnts.Add(em);
}
}
//
// construct the mRelationMentions matrix
//
ICollection <string> relKeys = doc.mRelationMentions.Keys;
foreach (string key_1 in relKeys)
{
AceRelationMention rm = doc.mRelationMentions[key_1];
sentence_1 = doc.mTokens[rm.GetArg(0).GetHead().GetTokenStart()].GetSentence();
//
// no need to adjust the number of rows: was done above
//
// store the relation mentions in increasing order
// (a) of the start position of their head, or
// (b) if start is the same, in increasing order of ends
List <AceRelationMention> sentRels = doc.mSentenceRelationMentions[sentence_1];
bool added = false;
for (int i = 0; i < sentRels.Count; i++)
{
AceRelationMention crt = sentRels[i];
if ((crt.GetMinTokenStart() > rm.GetMinTokenStart()) || (crt.GetMinTokenStart() == rm.GetMinTokenStart() && crt.GetMaxTokenEnd() > rm.GetMaxTokenEnd()))
{
sentRels.Add(i, rm);
added = true;
break;
}
}
if (!added)
{
sentRels.Add(rm);
}
}
//
// construct the mEventMentions matrix
//
ICollection <string> eventKeys = doc.mEventMentions.Keys;
foreach (string key_2 in eventKeys)
{
AceEventMention em = doc.mEventMentions[key_2];
sentence_1 = doc.mTokens[em.GetMinTokenStart()].GetSentence();
/*
* adjust the number of rows if necessary -- if you're wondering why we do
* this here again, (after we've done it for entities) it's because we can
* have an event with no entities near the end of the document and thus
* won't have created rows in mSentence*Mentions
*/
while (sentence_1 >= doc.mSentenceEntityMentions.Count)
{
doc.mSentenceEntityMentions.Add(new List <AceEntityMention>());
doc.mSentenceRelationMentions.Add(new List <AceRelationMention>());
doc.mSentenceEventMentions.Add(new List <AceEventMention>());
}
// store the event mentions in increasing order
// (a) first, event mentions with no arguments
// (b) then by the start position of their head, or
// (c) if start is the same, in increasing order of ends
List <AceEventMention> sentEvents = doc.mSentenceEventMentions[sentence_1];
bool added = false;
for (int i = 0; i < sentEvents.Count; i++)
{
AceEventMention crt = sentEvents[i];
if ((crt.GetMinTokenStart() > em.GetMinTokenStart()) || (crt.GetMinTokenStart() == em.GetMinTokenStart() && crt.GetMaxTokenEnd() > em.GetMaxTokenEnd()))
{
sentEvents.Add(i, em);
added = true;
break;
}
}
if (!added)
{
sentEvents.Add(em);
}
}
return(doc);
}