edu.nyu.jet.refres

Class CorefScorer

java.lang.Object

edu.nyu.jet.refres.DocumentScorer

edu.nyu.jet.refres.CorefScorer

public class CorefScorer
extends DocumentScorer

implements coreference scoring based on the metric developed by Marc Vilain for MUC-6. The two documents should have entity annotations, each representing (the referent of) a set of coreferential mentions. Each annotation should have a mentions feature whose value is a Vector of mentions -- the phrases referring to this entity. The scorer measures the overlap between the mention sets in the key and response documents.

Mentions in the response and key files are aligned based on the final character of the head of the mentions.

Field Summary

Fields

Modifier and Type	Field and Description
int	mappedMentionCount number of mentions in current response Document which have a corresponding mention in key Document (a mention whose head ends at the same position).
int	overallMappedMentionCount number of corresponding mentions in all documents processed by this instance of CorefScorer.
float	overallPrecision coreference link precison for all documents processed by this instance of CorefScorer.
float	overallRecall coreference link recall for all documents processed by this instance of CorefScorer.
float	precision coreference link precision for the most recently processed document.
float	recall coreference link recall for the most recently processed document.

Constructor Summary

Constructors

Constructor and Description
CorefScorer() create a new coreference scorer.

Method Summary

Methods

Modifier and Type	Method and Description
static Vector	findMentions(Document doc) return a Vector of all the mentions in the document (the union of the 'mentions' feature of all entities).
static void	main(String[] args)
void	report() writes to standard output a report on the most recently scored document.
void	score(Document responseDoc, Document keyDoc) compare the two documents, responseDoc and keyDoc, setting recall and precision.
void	summary() writes to standard output a report on the overall accuracy for all documents processed so far.

Methods inherited from class java.lang.Object

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

Field Detail

recall

public float recall

coreference link recall for the most recently processed document.

precision

public float precision

coreference link precision for the most recently processed document.

overallRecall

public float overallRecall

coreference link recall for all documents processed by this instance of CorefScorer.

overallPrecision

public float overallPrecision

coreference link precison for all documents processed by this instance of CorefScorer.

mappedMentionCount

public int mappedMentionCount

number of mentions in current response Document which have a corresponding mention in key Document (a mention whose head ends at the same position).

overallMappedMentionCount

public int overallMappedMentionCount

number of corresponding mentions in all documents processed by this instance of CorefScorer.

Constructor Detail

CorefScorer

public CorefScorer()

create a new coreference scorer.

Method Detail

score

public void score(Document responseDoc,
         Document keyDoc)

compare the two documents, responseDoc and keyDoc, setting recall and precision.

Specified by:

score in class DocumentScorer

findMentions

public static Vector findMentions(Document doc)

return a Vector of all the mentions in the document (the union of the 'mentions' feature of all entities).

report

public void report()

writes to standard output a report on the most recently scored document.

Specified by:

report in class DocumentScorer

summary

public void summary()

writes to standard output a report on the overall accuracy for all documents processed so far.

Specified by:

summary in class DocumentScorer

main

public static void main(String[] args)
                 throws IOException

Throws:

IOException