edu.stanford.nlp.dcoref
Class SieveCoreferenceSystem

java.lang.Object
  edu.stanford.nlp.dcoref.SieveCoreferenceSystem

public class SieveCoreferenceSystem
extends java.lang.Object

Multi-pass Sieve coreference resolution system (see EMNLP 2010 paper).

The main entry point for API is coref(Document document). The output is a map from CorefChain ID to corresponding CorefChain.

Author:

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

Nested Class Summary

static class

SieveCoreferenceSystem.LogFormatter

Field Summary

java.lang.String	conllMentionEvalScript Path for the official CoNLL scorer
int	currentSieve Current sieve index
java.util.List<Pair<java.lang.Integer,java.lang.Integer>>	linksCountInPass counter for links in passes (Pair)
static java.util.logging.Logger	logger
java.util.List<CorefScorer>	scoreBcubed
java.util.List<CorefScorer>	scoreMUC
java.util.List<CorefScorer>	scorePairwise Scores for each pass

Constructor Summary

SieveCoreferenceSystem(java.util.Properties props)

Method Summary

static boolean	checkClusters(java.util.logging.Logger logger, java.lang.String tag, Document document)
java.util.Map<java.lang.Integer,CorefChain>	coref(Document document) Extracts coreference clusters.
static void	debugPrintMentions(java.io.PrintStream out, java.lang.String tag, java.util.List<java.util.List<Mention>> mentions)
Dictionaries	dictionaries()
boolean	doScore()
static java.util.List<java.util.List<Mention>>	filterMentionsWithSingletonClusters(Document document, java.util.List<java.util.List<Mention>> mentions) Remove singleton clusters
static java.lang.String	formatPennTree(Tree parseTree) For printing tree in a better format
static java.lang.String	getConllEvalSummary(java.lang.String conllMentionEvalScript, java.lang.String goldFile, java.lang.String predictFile)
static java.util.List<Pair<IntTuple,IntTuple>>	getLinks(java.util.Map<java.lang.Integer,CorefChain> result)
void	initScorers()
static void	main(java.lang.String[] args) Needs the following properties: -props 'Location of coref.properties'
void	optimizeSieveOrdering(MentionExtractor mentionExtractor, java.util.Properties props, java.lang.String timestamp) Given a set of sieves, select an optimal ordering for the sieves by iterating over sieves, and selecting the one that gives the best score and adding sieves one at a time until no more sieves left
static void	printConllOutput(Document document, java.io.PrintWriter writer, boolean gold)
static void	printConllOutput(Document document, java.io.PrintWriter writer, boolean gold, boolean filterSingletons)
static void	printConllOutput(Document document, java.io.PrintWriter writer, java.util.List<java.util.List<Mention>> orderedMentions, boolean gold)
void	printF1(boolean printF1First)
protected static void	printList(java.util.logging.Logger logger, java.lang.String... args)
static void	printRawDoc(Document document, boolean gold) Print raw document for analysis
void	printTopK(java.util.logging.Logger logger, Document document, Semantics semantics) Print logs for error analysis
static double	runAndScoreCoref(SieveCoreferenceSystem corefSystem, MentionExtractor mentionExtractor, java.util.Properties props, java.lang.String timeStamp)
static void	runAndScoreCorefDist(java.lang.String runDistCmd, java.util.Properties props, java.lang.String propsFile) Run and score coref distributed
static void	runConllEval(java.lang.String conllMentionEvalScript, java.lang.String goldFile, java.lang.String predictFile, java.lang.String evalFile, java.lang.String errFile)
Semantics	semantics()
static java.lang.String	signature(java.util.Properties props)

Methods inherited from class java.lang.Object

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

Field Detail

logger

public static final java.util.logging.Logger logger

conllMentionEvalScript

public final java.lang.String conllMentionEvalScript

Path for the official CoNLL scorer

currentSieve

public int currentSieve

Current sieve index

linksCountInPass

public java.util.List<Pair<java.lang.Integer,java.lang.Integer>> linksCountInPass

counter for links in passes (Pair)

scorePairwise

public java.util.List<CorefScorer> scorePairwise

Scores for each pass

scoreBcubed

public java.util.List<CorefScorer> scoreBcubed

scoreMUC

public java.util.List<CorefScorer> scoreMUC

Constructor Detail

SieveCoreferenceSystem

public SieveCoreferenceSystem(java.util.Properties props)
                       throws java.lang.Exception

Throws:

java.lang.Exception

Method Detail

signature

public static java.lang.String signature(java.util.Properties props)

initScorers

public void initScorers()

doScore

public boolean doScore()

dictionaries

public Dictionaries dictionaries()

semantics

public Semantics semantics()

main

public static void main(java.lang.String[] args)
                 throws java.lang.Exception

Needs the following properties: -props 'Location of coref.properties'

Throws:

java.lang.Exception

runAndScoreCoref

public static double runAndScoreCoref(SieveCoreferenceSystem corefSystem,
                                      MentionExtractor mentionExtractor,
                                      java.util.Properties props,
                                      java.lang.String timeStamp)
                               throws java.lang.Exception

Throws:

java.lang.Exception

runAndScoreCorefDist

public static void runAndScoreCorefDist(java.lang.String runDistCmd,
                                        java.util.Properties props,
                                        java.lang.String propsFile)
                                 throws java.lang.Exception

Run and score coref distributed

Throws:

java.lang.Exception

optimizeSieveOrdering

public void optimizeSieveOrdering(MentionExtractor mentionExtractor,
                                  java.util.Properties props,
                                  java.lang.String timestamp)
                           throws java.lang.Exception

Given a set of sieves, select an optimal ordering for the sieves by iterating over sieves, and selecting the one that gives the best score and adding sieves one at a time until no more sieves left

Throws:

java.lang.Exception

coref

public java.util.Map<java.lang.Integer,CorefChain> coref(Document document)
                                                  throws java.lang.Exception

Extracts coreference clusters. This is the main API entry point for coreference resolution. Return a map from CorefChain ID to corresponding CorefChain.

Throws:

java.lang.Exception

filterMentionsWithSingletonClusters

public static java.util.List<java.util.List<Mention>> filterMentionsWithSingletonClusters(Document document,
                                                                                          java.util.List<java.util.List<Mention>> mentions)

Remove singleton clusters

runConllEval

public static void runConllEval(java.lang.String conllMentionEvalScript,
                                java.lang.String goldFile,
                                java.lang.String predictFile,
                                java.lang.String evalFile,
                                java.lang.String errFile)
                         throws java.io.IOException

Throws:

java.io.IOException

getConllEvalSummary

public static java.lang.String getConllEvalSummary(java.lang.String conllMentionEvalScript,
                                                   java.lang.String goldFile,
                                                   java.lang.String predictFile)
                                            throws java.io.IOException

Throws:

java.io.IOException

printTopK

public void printTopK(java.util.logging.Logger logger,
                      Document document,
                      Semantics semantics)

Print logs for error analysis

printF1

public void printF1(boolean printF1First)

printList

protected static void printList(java.util.logging.Logger logger,
                                java.lang.String... args)

formatPennTree

public static java.lang.String formatPennTree(Tree parseTree)

For printing tree in a better format

printConllOutput

public static void printConllOutput(Document document,
                                    java.io.PrintWriter writer,
                                    boolean gold)

printConllOutput

public static void printConllOutput(Document document,
                                    java.io.PrintWriter writer,
                                    boolean gold,
                                    boolean filterSingletons)

printConllOutput

public static void printConllOutput(Document document,
                                    java.io.PrintWriter writer,
                                    java.util.List<java.util.List<Mention>> orderedMentions,
                                    boolean gold)

printRawDoc

public static void printRawDoc(Document document,
                               boolean gold)
                        throws java.io.FileNotFoundException

Print raw document for analysis

Throws:

java.io.FileNotFoundException

getLinks

public static java.util.List<Pair<IntTuple,IntTuple>> getLinks(java.util.Map<java.lang.Integer,CorefChain> result)

debugPrintMentions

public static void debugPrintMentions(java.io.PrintStream out,
                                      java.lang.String tag,
                                      java.util.List<java.util.List<Mention>> mentions)

checkClusters

public static boolean checkClusters(java.util.logging.Logger logger,
                                    java.lang.String tag,
                                    Document document)