edu.nyu.jet.aceJet

Class EventTagger

java.lang.Object

edu.nyu.jet.aceJet.EventTagger

public class EventTagger
extends Object

assigns ACE events to a Document, given the entities, times, and values.

This is a corpus-trained tagger which uses a dual-pronged approach to identifying events and their arguments.

One approach is based on creating a pattern from each event mention in the training corpus. The pattern encompasses the event trigger and all of its arguments, recording the connection between the trigger and the arguments either as a sequence of chunks or as a path in a syntax graph. At decoding time, a partial match of this pattern against the input can trigger an event.

The other approach is based on a set of maximum entropy classifiers. An argument classifier decides whether a given entity mention is an argument of a trigger, and a separate role classifier decides which role this argument takes. After all potential arguments have been classified, an event classifier decides whether, based on the arguments found, an event should be triggered.

Constructor Summary

Constructors

Constructor and Description
EventTagger()

Method Summary

Methods

Modifier and Type	Method and Description
void	acquirePatterns(Document doc, AceDocument aceDoc, String docId) trains the tagger from document 'doc' and corresponding AceDocument (APF file) aceDoc.
void	evaluatePatterns(Document doc, AceDocument aceDoc, String docId) applies the learned patterns to Document 'doc' and records the number of times it produced a correct or incorrect event.
void	eventCoref(AceDocument aceDoc, Document doc, SyntacticRelationSet relations) performs coreference on the events in an Ace document.
void	load(String fileName) load an event pattern set in the form saved by the 'save' method.
void	loadAllModels(String modelDir)
static void	main(String[] args) tag a set of documents with ACE event information, using entity and value information from an existing APF file, typically the key.
void	report(String reportFile) write a human-readable report of all the event patterns to file reportFile.
void	save(String fileName) write out all the patterns to file 'fileName' in a form which can be reloaded by the 'load' method.
void	tag(Document doc, AceDocument aceDoc, String currentDocPath, String docId) identify ACE events in Document 'doc' and add them to 'aceDoc'.
void	tag(String fileList) add event information to the APF files of the documents on fileList.

Methods inherited from class java.lang.Object

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

Constructor Detail

EventTagger

public EventTagger()

Method Detail

acquirePatterns

public void acquirePatterns(Document doc,
                   AceDocument aceDoc,
                   String docId)

trains the tagger from document 'doc' and corresponding AceDocument (APF file) aceDoc.

evaluatePatterns

public void evaluatePatterns(Document doc,
                    AceDocument aceDoc,
                    String docId)

applies the learned patterns to Document 'doc' and records the number of times it produced a correct or incorrect event.

tag

public void tag(Document doc,
       AceDocument aceDoc,
       String currentDocPath,
       String docId)

identify ACE events in Document 'doc' and add them to 'aceDoc'.

tag

public void tag(String fileList)
         throws IOException

add event information to the APF files of the documents on fileList.

Throws:

IOException

eventCoref

public void eventCoref(AceDocument aceDoc,
              Document doc,
              SyntacticRelationSet relations)

performs coreference on the events in an Ace document. On entry, the AceDocument aceDoc should have a set of events each with a single mention. The event mentions which are believed to corefer are combined into a single event.

report

public void report(String reportFile)
            throws IOException

write a human-readable report of all the event patterns to file reportFile.

Throws:

IOException

save

public void save(String fileName)
          throws IOException

write out all the patterns to file 'fileName' in a form which can be reloaded by the 'load' method.

Throws:

IOException

load

public void load(String fileName)
          throws IOException

load an event pattern set in the form saved by the 'save' method.

Throws:

IOException

loadAllModels

public void loadAllModels(String modelDir)
                   throws IOException

Throws:

IOException

main

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

tag a set of documents with ACE event information, using entity and value information from an existing APF file, typically the key. Takes the following arguments:

• props: Jet property file
• fileList: list of files to be processed (one per line)
• docDir: directory containing text and APF files
• modelDir: directory containing event patterns and models
• outDir: directory to receive updated APF files including event information
• glarfDir: directory containing glarf triples (optional)
• glarfSuffix: file extension for glarf files

Throws:

IOException