edu.stanford.nlp.ie.machinereading
Class GenericDataSetReader

java.lang.Object
  edu.stanford.nlp.ie.machinereading.GenericDataSetReader

Direct Known Subclasses:

AceReader

public class GenericDataSetReader
extends java.lang.Object

Author:

Andrey Gusev, Mihai

Field Summary

protected boolean	calculateHeadSpan If true, sets the head span to match the syntactic head of the extent.
protected boolean	forceGenerationOfIndexSpans If true, it regenerates the index spans for all tree nodes (useful for KBP)
protected HeadFinder	headFinder Finds the syntactic head of a syntactic constituent
protected java.util.logging.Logger	logger
protected Annotator	parserProcessor Additional NL processor that implements only syntactic parsing (needed for head detection) We need this processor to detect heads of predicted entities that cannot be matched to an existing constituent.
protected boolean	preProcessSentences If true, we perform syntactic analysis of the dataset sentences and annotations
protected StanfordCoreNLP	processor NL processor to use for sentence pre-processing
protected boolean	useNewHeadFinder Only around for legacy results

Constructor Summary

GenericDataSetReader()

GenericDataSetReader(StanfordCoreNLP processor, boolean preProcessSentences, boolean calculateHeadSpan, boolean forceGenerationOfIndexSpans)

Method Summary

int	assignSyntacticHead(EntityMention ent, Tree tree, java.util.List<CoreLabel> tokens, boolean setHeadSpan) Find the index of the head of an entity.
static void	convertToCoreLabels(Tree tree) Converts the tree labels to CoreLabels.
Tree	findSyntacticHead(EntityMention ent, Tree root, java.util.List<CoreLabel> tokens) Finds the syntactic head of the given entity mention.
java.util.logging.Level	getLoggerLevel()
Annotator	getParser()
Tree	originalFindSyntacticHead(EntityMention ent, Tree root, java.util.List<CoreLabel> tokens) This is the original version of findSyntacticHead(edu.stanford.nlp.ie.machinereading.structure.EntityMention, edu.stanford.nlp.trees.Tree, java.util.List) before Chris's modifications.
protected Tree	parse(java.util.List<CoreLabel> tokens)
protected Tree	parse(java.util.List<CoreLabel> tokens, java.util.List<ParserConstraint> constraints)
Annotation	parse(java.lang.String path) Parses one file or directory with data from one domain
protected Tree	parseStrings(java.util.List<java.lang.String> tokens)
void	preProcessSentences(Annotation dataset) Take a dataset Annotation, generate their parse trees and identify syntactic heads (and head spans, if necessary)
Annotation	read(java.lang.String path)
void	setLoggerLevel(java.util.logging.Level level)
void	setProcessor(StanfordCoreNLP p)
void	setUseNewHeadFinder(boolean useNewHeadFinder)

Methods inherited from class java.lang.Object

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

Field Detail

logger

protected java.util.logging.Logger logger

headFinder

protected final HeadFinder headFinder

Finds the syntactic head of a syntactic constituent

processor

protected StanfordCoreNLP processor

NL processor to use for sentence pre-processing

parserProcessor

protected Annotator parserProcessor

Additional NL processor that implements only syntactic parsing (needed for head detection) We need this processor to detect heads of predicted entities that cannot be matched to an existing constituent. This is created on demand, only when necessary

preProcessSentences

protected final boolean preProcessSentences

If true, we perform syntactic analysis of the dataset sentences and annotations

calculateHeadSpan

protected final boolean calculateHeadSpan

If true, sets the head span to match the syntactic head of the extent. Otherwise, the head span is not modified. This is enabled for the NFL domain, where head spans are not given.

forceGenerationOfIndexSpans

protected final boolean forceGenerationOfIndexSpans

If true, it regenerates the index spans for all tree nodes (useful for KBP)

useNewHeadFinder

protected boolean useNewHeadFinder

Only around for legacy results

Constructor Detail

GenericDataSetReader

public GenericDataSetReader()

GenericDataSetReader

public GenericDataSetReader(StanfordCoreNLP processor,
                            boolean preProcessSentences,
                            boolean calculateHeadSpan,
                            boolean forceGenerationOfIndexSpans)

Method Detail

setProcessor

public void setProcessor(StanfordCoreNLP p)

setUseNewHeadFinder

public void setUseNewHeadFinder(boolean useNewHeadFinder)

getParser

public Annotator getParser()

setLoggerLevel

public void setLoggerLevel(java.util.logging.Level level)

getLoggerLevel

public java.util.logging.Level getLoggerLevel()

parse

public final Annotation parse(java.lang.String path)
                       throws java.io.IOException

Parses one file or directory with data from one domain

Parameters:

path -

Throws:

java.io.IOException

read

public Annotation read(java.lang.String path)
                throws java.lang.Exception

Throws:

java.lang.Exception

assignSyntacticHead

public int assignSyntacticHead(EntityMention ent,
                               Tree tree,
                               java.util.List<CoreLabel> tokens,
                               boolean setHeadSpan)

Find the index of the head of an entity.

Parameters:

ent - The entity mention

tree - The Tree for the entire sentence in which it occurs.

tokens - The Sentence in which it occurs

setHeadSpan - Whether to set the head span in the entity mention.

Returns:

The index of the entity head

preProcessSentences

public void preProcessSentences(Annotation dataset)

Take a dataset Annotation, generate their parse trees and identify syntactic heads (and head spans, if necessary)

convertToCoreLabels

public static void convertToCoreLabels(Tree tree)

Converts the tree labels to CoreLabels. We need this because we store additional info in the CoreLabel, like token span.

Parameters:

tree -

findSyntacticHead

public Tree findSyntacticHead(EntityMention ent,
                              Tree root,
                              java.util.List<CoreLabel> tokens)

Finds the syntactic head of the given entity mention.

Parameters:

ent - The entity mention

root - The Tree for the entire sentence in which it occurs.

tokens - The Sentence in which it occurs

Returns:

The tree object corresponding to the head. This MUST be a child of root. It will be a leaf in the parse tree.

originalFindSyntacticHead

public Tree originalFindSyntacticHead(EntityMention ent,
                                      Tree root,
                                      java.util.List<CoreLabel> tokens)

This is the original version of findSyntacticHead(edu.stanford.nlp.ie.machinereading.structure.EntityMention, edu.stanford.nlp.trees.Tree, java.util.List) before Chris's modifications. There's no good reason to use it except for producing historical results. It Finds the syntactic head of the given entity mention.

Parameters:

ent - The entity mention

root - The Tree for the entire sentence in which it occurs.

tokens - The Sentence in which it occurs

Returns:

The tree object corresponding to the head. This MUST be a child of root. It will be a leaf in the parse tree.

parseStrings

protected Tree parseStrings(java.util.List<java.lang.String> tokens)

parse

protected Tree parse(java.util.List<CoreLabel> tokens)

parse

protected Tree parse(java.util.List<CoreLabel> tokens,
                     java.util.List<ParserConstraint> constraints)