/*
* Sets the head word and the index for an entity, given the parse tree for
* the sentence containing the entity.
*
* This code is no longer used, but I've kept it around (at least for now) as
* reference when we modify preProcessSentences().
*/
private void SetHeadWord(EntityMention entity, Tree tree)
{
IList <Tree> leaves = tree.GetLeaves();
Tree argRoot = tree.JoinNode(leaves[entity.GetExtentTokenStart()], leaves[entity.GetExtentTokenEnd()]);
Tree headWordNode = argRoot.HeadTerminal(headFinder);
int headWordIndex = GetIndexByObjectEquality(leaves, headWordNode);
if (StringUtils.IsPunct(leaves[entity.GetExtentTokenEnd()].Label().Value().Trim()) && (headWordIndex >= entity.GetExtentTokenEnd() || headWordIndex < entity.GetExtentTokenStart()))
{
argRoot = tree.JoinNode(leaves[entity.GetExtentTokenStart()], leaves[entity.GetExtentTokenEnd() - 1]);
headWordNode = argRoot.HeadTerminal(headFinder);
headWordIndex = GetIndexByObjectEquality(leaves, headWordNode);
if (headWordIndex >= entity.GetExtentTokenStart() && headWordIndex <= entity.GetExtentTokenEnd() - 1)
{
entity.SetHeadTokenPosition(headWordIndex);
entity.SetHeadTokenSpan(new Span(headWordIndex, headWordIndex + 1));
}
}
if (headWordIndex >= entity.GetExtentTokenStart() && headWordIndex <= entity.GetExtentTokenEnd())
{
entity.SetHeadTokenPosition(headWordIndex);
entity.SetHeadTokenSpan(new Span(headWordIndex, headWordIndex + 1));
}
else
{
// Re-parse the argument words by themselves
// Get the list of words in the arg by looking at the leaves between
// arg.getExtentTokenStart() and arg.getExtentTokenEnd() inclusive
IList <string> argWords = new List <string>();
for (int i = entity.GetExtentTokenStart(); i <= entity.GetExtentTokenEnd(); i++)
{
argWords.Add(leaves[i].Label().Value());
}
if (StringUtils.IsPunct(argWords[argWords.Count - 1]))
{
argWords.Remove(argWords.Count - 1);
}
Tree argTree = ParseStrings(argWords);
headWordNode = argTree.HeadTerminal(headFinder);
headWordIndex = GetIndexByObjectEquality(argTree.GetLeaves(), headWordNode) + entity.GetExtentTokenStart();
entity.SetHeadTokenPosition(headWordIndex);
entity.SetHeadTokenSpan(new Span(headWordIndex, headWordIndex + 1));
}
}
/// <summary>
/// This is the original version of
/// <see cref="FindSyntacticHead(Edu.Stanford.Nlp.IE.Machinereading.Structure.EntityMention, Edu.Stanford.Nlp.Trees.Tree, System.Collections.Generic.IList{E})"/>
/// before Chris's modifications.
/// There's no good reason to use it except for producing historical results.
/// It Finds the syntactic head of the given entity mention.
/// </summary>
/// <param name="ent">The entity mention</param>
/// <param name="root">The Tree for the entire sentence in which it occurs.</param>
/// <param name="tokens">The Sentence in which it occurs</param>
/// <returns>
/// The tree object corresponding to the head. This MUST be a child of root.
/// It will be a leaf in the parse tree.
/// </returns>
public virtual Tree OriginalFindSyntacticHead(EntityMention ent, Tree root, IList <CoreLabel> tokens)
{
logger.Fine("Searching for tree matching " + ent);
Tree exactMatch = FindTreeWithSpan(root, ent.GetExtentTokenStart(), ent.GetExtentTokenEnd());
//
// found an exact match
//
if (exactMatch != null)
{
logger.Fine("Mention \"" + ent + "\" mapped to tree: " + PrintTree(exactMatch));
return(SafeHead(exactMatch));
}
//
// no exact match found
// in this case, we parse the actual extent of the mention
//
IList <CoreLabel> extentTokens = new List <CoreLabel>();
for (int i = ent.GetExtentTokenStart(); i < ent.GetExtentTokenEnd(); i++)
{
extentTokens.Add(tokens[i]);
}
Tree tree = Parse(extentTokens);
logger.Fine("No exact match found. Local parse:\n" + tree.PennString());
ConvertToCoreLabels(tree);
tree.IndexSpans(ent.GetExtentTokenStart());
Tree extentHead = SafeHead(tree);
System.Diagnostics.Debug.Assert((extentHead != null));
// extentHead is a child in the local extent parse tree. we need to find the
// corresponding node in the main tree
CoreLabel l = (CoreLabel)extentHead.Label();
Tree realHead = FindTreeWithSpan(root, l.Get(typeof(CoreAnnotations.BeginIndexAnnotation)), l.Get(typeof(CoreAnnotations.EndIndexAnnotation)));
System.Diagnostics.Debug.Assert((realHead != null));
return(realHead);
}
/// <summary>Finds the syntactic head of the given entity mention.</summary>
/// <param name="ent">The entity mention</param>
/// <param name="root">The Tree for the entire sentence in which it occurs.</param>
/// <param name="tokens">The Sentence in which it occurs</param>
/// <returns>
/// The tree object corresponding to the head. This MUST be a child of root.
/// It will be a leaf in the parse tree.
/// </returns>
public virtual Tree FindSyntacticHead(EntityMention ent, Tree root, IList <CoreLabel> tokens)
{
if (!useNewHeadFinder)
{
return(OriginalFindSyntacticHead(ent, root, tokens));
}
logger.Fine("Searching for tree matching " + ent);
Tree exactMatch = FindTreeWithSpan(root, ent.GetExtentTokenStart(), ent.GetExtentTokenEnd());
//
// found an exact match
//
if (exactMatch != null)
{
logger.Fine("Mention \"" + ent + "\" mapped to tree: " + PrintTree(exactMatch));
return(SafeHead(exactMatch));
}
// no exact match found
// in this case, we parse the actual extent of the mention, embedded in a sentence
// context, so as to make the parser work better :-)
int approximateness = 0;
IList <CoreLabel> extentTokens = new List <CoreLabel>();
extentTokens.Add(InitCoreLabel("It"));
extentTokens.Add(InitCoreLabel("was"));
int AddedWords = 2;
for (int i = ent.GetExtentTokenStart(); i < ent.GetExtentTokenEnd(); i++)
{
// Add everything except separated dashes! The separated dashes mess with the parser too badly.
CoreLabel label = tokens[i];
if (!"-".Equals(label.Word()))
{
extentTokens.Add(tokens[i]);
}
else
{
approximateness++;
}
}
extentTokens.Add(InitCoreLabel("."));
// constrain the parse to the part we're interested in.
// Starting from ADDED_WORDS comes from skipping "It was".
// -1 to exclude the period.
// We now let it be any kind of nominal constituent, since there
// are VP and S ones
ParserConstraint constraint = new ParserConstraint(AddedWords, extentTokens.Count - 1, ".*");
IList <ParserConstraint> constraints = Java.Util.Collections.SingletonList(constraint);
Tree tree = Parse(extentTokens, constraints);
logger.Fine("No exact match found. Local parse:\n" + tree.PennString());
ConvertToCoreLabels(tree);
tree.IndexSpans(ent.GetExtentTokenStart() - AddedWords);
// remember it has ADDED_WORDS extra words at the beginning
Tree subtree = FindPartialSpan(tree, ent.GetExtentTokenStart());
Tree extentHead = SafeHead(subtree);
logger.Fine("Head is: " + extentHead);
System.Diagnostics.Debug.Assert((extentHead != null));
// extentHead is a child in the local extent parse tree. we need to find the corresponding node in the main tree
// Because we deleted dashes, it's index will be >= the index in the extent parse tree
CoreLabel l = (CoreLabel)extentHead.Label();
// Tree realHead = findTreeWithSpan(root, l.get(CoreAnnotations.BeginIndexAnnotation.class), l.get(CoreAnnotations.EndIndexAnnotation.class));
Tree realHead = FunkyFindLeafWithApproximateSpan(root, l.Value(), l.Get(typeof(CoreAnnotations.BeginIndexAnnotation)), approximateness);
if (realHead != null)
{
logger.Fine("Chosen head: " + realHead);
}
return(realHead);
}
private Annotation ReadSentence(string docId, IEnumerator <string> lineIterator)
{
Annotation sentence = new Annotation(string.Empty);
sentence.Set(typeof(CoreAnnotations.DocIDAnnotation), docId);
sentence.Set(typeof(MachineReadingAnnotations.EntityMentionsAnnotation), new List <EntityMention>());
// we'll need to set things like the tokens and textContent after we've
// fully read the sentence
// contains the full text that we've read so far
StringBuilder textContent = new StringBuilder();
int tokenCount = 0;
// how many tokens we've seen so far
IList <CoreLabel> tokens = new List <CoreLabel>();
// when we've seen two blank lines in a row, this sentence is over (one
// blank line separates the sentence and the relations
int numBlankLinesSeen = 0;
string sentenceID = null;
// keeps tracks of entities we've seen so far for use by relations
IDictionary <string, EntityMention> indexToEntityMention = new Dictionary <string, EntityMention>();
while (lineIterator.MoveNext() && numBlankLinesSeen < 2)
{
string currentLine = lineIterator.Current;
currentLine = currentLine.Replace("COMMA", ",");
IList <string> pieces = StringUtils.Split(currentLine);
string identifier;
int size = pieces.Count;
switch (size)
{
case 1:
{
// blank line between sentences or relations
numBlankLinesSeen++;
break;
}
case 3:
{
// relation
string type = pieces[2];
IList <ExtractionObject> args = new List <ExtractionObject>();
EntityMention entity1 = indexToEntityMention[pieces[0]];
EntityMention entity2 = indexToEntityMention[pieces[1]];
args.Add(entity1);
args.Add(entity2);
Span span = new Span(entity1.GetExtentTokenStart(), entity2.GetExtentTokenEnd());
// identifier = "relation" + sentenceID + "-" + sentence.getAllRelations().size();
identifier = RelationMention.MakeUniqueId();
RelationMention relationMention = new RelationMention(identifier, sentence, span, type, null, args);
AnnotationUtils.AddRelationMention(sentence, relationMention);
break;
}
case 9:
{
// token
/*
* Roth token lines look like this:
*
* 19 Peop 9 O NNP/NNP Jamal/Ghosheh O O O
*/
// Entities may be multiple words joined by '/'; we split these up
IList <string> words = StringUtils.Split(pieces[5], "/");
//List<String> postags = StringUtils.split(pieces.get(4),"/");
string text = StringUtils.Join(words, " ");
identifier = "entity" + pieces[0] + '-' + pieces[2];
string nerTag = GetNormalizedNERTag(pieces[1]);
// entity type of the word/expression
if (sentenceID == null)
{
sentenceID = pieces[0];
}
if (!nerTag.Equals("O"))
{
Span extentSpan = new Span(tokenCount, tokenCount + words.Count);
// Temporarily sets the head span to equal the extent span.
// This is so the entity has a head (in particular, getValue() works) even if preprocessSentences isn't called.
// The head span is later modified if preprocessSentences is called.
EntityMention entity = new EntityMention(identifier, sentence, extentSpan, extentSpan, nerTag, null, null);
AnnotationUtils.AddEntityMention(sentence, entity);
// we can get by using these indices as strings since we only use them
// as a hash key
string index = pieces[2];
indexToEntityMention[index] = entity;
}
// int i =0;
foreach (string word in words)
{
CoreLabel label = new CoreLabel();
label.SetWord(word);
//label.setTag(postags.get(i));
label.Set(typeof(CoreAnnotations.TextAnnotation), word);
label.Set(typeof(CoreAnnotations.ValueAnnotation), word);
// we don't set TokenBeginAnnotation or TokenEndAnnotation since we're
// not keeping track of character offsets
tokens.Add(label);
}
// i++;
textContent.Append(text);
textContent.Append(' ');
tokenCount += words.Count;
break;
}
}
}
sentence.Set(typeof(CoreAnnotations.TextAnnotation), textContent.ToString());
sentence.Set(typeof(CoreAnnotations.ValueAnnotation), textContent.ToString());
sentence.Set(typeof(CoreAnnotations.TokensAnnotation), tokens);
sentence.Set(typeof(CoreAnnotations.SentenceIDAnnotation), sentenceID);
return(sentence);
}