edu.stanford.nlp.dcoref

Class MentionExtractor

java.lang.Object

edu.stanford.nlp.dcoref.MentionExtractor

Direct Known Subclasses:

ACEMentionExtractor, CoNLLMentionExtractor, MUCMentionExtractor

public class MentionExtractor
extends java.lang.Object

Generic mention extractor from a corpus.

Author:

Jenny Finkel, Mihai Surdeanu, Karthik Raghunathan, Heeyoung Lee, Sudarshan Rangarajan

Field Summary

Fields

Modifier and Type	Field and Description
protected java.lang.String	currentDocumentID
protected Dictionaries	dictionaries
protected int	maxID The maximum mention ID: for preventing duplicated mention ID assignment
CorefMentionFinder	mentionFinder
protected Semantics	semantics
protected LogisticClassifier<java.lang.String,java.lang.String>	singletonPredictor
protected StanfordCoreNLP	stanfordProcessor
static boolean	VERBOSE

Constructor Summary

Constructors

Constructor and Description
MentionExtractor(Dictionaries dict, Semantics semantics)

Method Summary

Modifier and Type	Method and Description
Document	arrange(Annotation anno, java.util.List<java.util.List<CoreLabel>> words, java.util.List<Tree> trees, java.util.List<java.util.List<Mention>> unorderedMentions)
java.util.List<java.util.List<Mention>>	arrange(Annotation anno, java.util.List<java.util.List<CoreLabel>> words, java.util.List<Tree> trees, java.util.List<java.util.List<Mention>> unorderedMentions, boolean doMergeLabels) Post-processes the extracted mentions.
Document	arrange(Annotation anno, java.util.List<java.util.List<CoreLabel>> words, java.util.List<Tree> trees, java.util.List<java.util.List<Mention>> unorderedMentions, java.util.List<java.util.List<Mention>> unorderedGoldMentions, boolean doMergeLabels)
static Tree	findExactMatch(Tree tree, int first, int last) Finds the tree the matches this span exactly
protected int	getHeadIndex(Tree t)
static void	initializeUtterance(java.util.List<CoreLabel> tokens)
protected static StanfordCoreNLP	loadStanfordProcessor(java.util.Properties props) Load Stanford Processor: skip unnecessary annotator
static void	mergeLabels(Tree tree, java.util.List<CoreLabel> sentence) Sets the label of the leaf nodes of a Tree to be the CoreLabels in the given sentence.
Document	nextDoc() Extracts the info relevant for coref from the next document in the corpus
void	resetDocs() Reset so that we start at the beginning of the document collection
void	setMentionFinder(CorefMentionFinder mentionFinder)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

currentDocumentID

protected java.lang.String currentDocumentID

dictionaries

protected final Dictionaries dictionaries

semantics

protected final Semantics semantics

mentionFinder

public CorefMentionFinder mentionFinder

stanfordProcessor

protected StanfordCoreNLP stanfordProcessor

singletonPredictor

protected LogisticClassifier<java.lang.String,java.lang.String> singletonPredictor

maxID

protected int maxID

The maximum mention ID: for preventing duplicated mention ID assignment

VERBOSE

public static final boolean VERBOSE

See Also:

Constant Field Values

Constructor Detail

MentionExtractor

public MentionExtractor(Dictionaries dict,
                        Semantics semantics)

Method Detail

setMentionFinder

public void setMentionFinder(CorefMentionFinder mentionFinder)

nextDoc

public Document nextDoc()
                 throws java.lang.Exception

Extracts the info relevant for coref from the next document in the corpus

Returns:

List of mentions found in each sentence ordered according to the tree traversal.

Throws:

java.lang.Exception

resetDocs

public void resetDocs()

Reset so that we start at the beginning of the document collection

arrange

public Document arrange(Annotation anno,
                        java.util.List<java.util.List<CoreLabel>> words,
                        java.util.List<Tree> trees,
                        java.util.List<java.util.List<Mention>> unorderedMentions)
                 throws java.lang.Exception

Throws:

java.lang.Exception

getHeadIndex

protected int getHeadIndex(Tree t)

arrange

public Document arrange(Annotation anno,
                        java.util.List<java.util.List<CoreLabel>> words,
                        java.util.List<Tree> trees,
                        java.util.List<java.util.List<Mention>> unorderedMentions,
                        java.util.List<java.util.List<Mention>> unorderedGoldMentions,
                        boolean doMergeLabels)
                 throws java.lang.Exception

Throws:

java.lang.Exception

arrange

public java.util.List<java.util.List<Mention>> arrange(Annotation anno,
                                                       java.util.List<java.util.List<CoreLabel>> words,
                                                       java.util.List<Tree> trees,
                                                       java.util.List<java.util.List<Mention>> unorderedMentions,
                                                       boolean doMergeLabels)
                                                throws java.lang.Exception

Post-processes the extracted mentions. Here we set the Mention fields required for coref and order mentions by tree-traversal order.

Parameters:

words - List of words in each sentence, in textual order

trees - List of trees, one per sentence

unorderedMentions - List of unordered, unprocessed mentions Each mention MUST have startIndex and endIndex set! Optionally, if scoring is desired, mentions must have mentionID and originalRef set. All the other Mention fields are set here.

Returns:

List of mentions ordered according to the tree traversal

Throws:

java.lang.Exception

mergeLabels

public static void mergeLabels(Tree tree,
                               java.util.List<CoreLabel> sentence)

Sets the label of the leaf nodes of a Tree to be the CoreLabels in the given sentence. The original value() of the Tree nodes is preserved, and otherwise the label of tree leaves becomes the label from the List.

findExactMatch

public static Tree findExactMatch(Tree tree,
                                  int first,
                                  int last)

Finds the tree the matches this span exactly

Parameters:

tree - Leaves must be indexed!

first - First element in the span (first position has offset 1)

last - Last element included in the span (first position has offset 1)

loadStanfordProcessor

protected static StanfordCoreNLP loadStanfordProcessor(java.util.Properties props)

Load Stanford Processor: skip unnecessary annotator

initializeUtterance

public static void initializeUtterance(java.util.List<CoreLabel> tokens)