edu.nyu.jet.aceJet

Class RelationTagger

java.lang.Object

edu.nyu.jet.aceJet.RelationTagger

public class RelationTagger
extends Object

procedures for training the ACE relation tagger and for tagging new text. Currently the tagger uses a set of 4 maximum entropy models:

1. a detection model for relations within base noun phrases
2. a classification model for relations within base noun phrases
3. a detection model for relations between other entity pairs
4. a classification model for relations between other entity pairs

The four feature files and four model files are stored together in a single model directory. This directory may also have an "argTypes" file, specifying the allowed argument types for each relation type and subtype.

This relation tagger is based on maximum entropy models and is a much faster replacement for the nearest-neighbor model used until 2011.

Field Summary

Fields

Modifier and Type	Field and Description
static boolean	mergeMultipleRelations if true, multiple relation mentions (with different types) between a single pair of entities are merged (coref'ed) into a single relation, with the type of the first mention.

Constructor Summary

Constructors

Constructor and Description
RelationTagger()

Method Summary

Methods

Modifier and Type	Method and Description
static void	findRelations(String currentDoc, Document d, AceDocument ad) relation 'decoder': using previously trained models, identifies the relations in document 'doc' (from file name 'currentDoc') and adds them as AceRelations to AceDocument 'aceDoc'.
static void	main(String[] args) trains a new relation tagger.

Methods inherited from class java.lang.Object

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

Field Detail

mergeMultipleRelations

public static boolean mergeMultipleRelations

if true, multiple relation mentions (with different types) between a single pair of entities are merged (coref'ed) into a single relation, with the type of the first mention.

Constructor Detail

RelationTagger

public RelationTagger()

Method Detail

main

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

trains a new relation tagger. Takes 5n+2 arguments:
RelationTagger configFile [docList textDir textSuffix apfDir apfSuffix] modelDirectory
where arguments 2 through 6 are repeated for each training corpus (training data from multiple corpora can be combined into a single model). The resulting maxent models are written into 'modelDirectory'.

Throws:

IOException

findRelations

public static void findRelations(String currentDoc,
                 Document d,
                 AceDocument ad)

relation 'decoder': using previously trained models, identifies the relations in document 'doc' (from file name 'currentDoc') and adds them as AceRelations to AceDocument 'aceDoc'.