info.ephyra.questionanalysis.atype
Class FocusFinder

java.lang.Object
  info.ephyra.questionanalysis.atype.FocusFinder

public class FocusFinder
extends java.lang.Object

Finder of the focus word, or target word, of a question.

Version:

2008-02-10

Author:

Justin Betteridge

Field Summary

private static boolean	isInitialized
private static org.apache.log4j.Logger	log
private static java.util.List<edu.cmu.lti.chineseNLP.util.Tree>	treeTemplates
private static java.lang.String	treeTemplatesFile
private static java.lang.String[]	whLabels

Constructor Summary

FocusFinder()

Method Summary

static edu.cmu.lti.chineseNLP.util.Tree	findFocusNode(edu.cmu.lti.chineseNLP.util.Tree tree) Finds the focus, or target, word of a question using the specified tree templates and default rules that look for question words.
static edu.cmu.lti.javelin.qa.Term	findFocusTerm(java.util.List<edu.cmu.lti.javelin.qa.Term> terms) Given a list of Terms, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.
static edu.cmu.lti.javelin.qa.Term	findFocusTerm(java.lang.String question) Given a sentence, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.
static edu.cmu.lti.javelin.qa.Term	findFocusTerm(edu.cmu.lti.chineseNLP.util.Tree tree) Finds the focus word, given a Tree.
static java.lang.String	findFocusWord(java.util.List<edu.cmu.lti.javelin.qa.Term> terms) Given a list of Terms, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.
static java.lang.String	findFocusWord(java.lang.String question) Given a sentence, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.
static java.lang.String	findFocusWord(edu.cmu.lti.chineseNLP.util.Tree tree) Finds the focus word, given a Tree.
static edu.cmu.lti.chineseNLP.util.Tree	getHeadWordOrPhrase(edu.cmu.lti.chineseNLP.util.Tree tree) Extracts the head word or phrase from the given Tree node, which is assumed to be an NP.
static void	initialize() Initializes static resources.
static boolean	isInitialized()
static void	main(java.lang.String[] args) Extracts and prints out the focus word for each question, given a file of questions.
static void	setInitialized(boolean isInitialized)

Methods inherited from class java.lang.Object

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

Field Detail

log

private static org.apache.log4j.Logger log

isInitialized

private static boolean isInitialized

treeTemplates

private static java.util.List<edu.cmu.lti.chineseNLP.util.Tree> treeTemplates

treeTemplatesFile

private static java.lang.String treeTemplatesFile

whLabels

private static java.lang.String[] whLabels

Constructor Detail

FocusFinder

public FocusFinder()

Method Detail

initialize

public static void initialize()
                       throws java.lang.Exception

Initializes static resources. The input properties which must be defined are:

• edu.cmu.lti.javelin.qa.english.FocusFinder.treeTemplatesFile : the location of a file containing tree templates to use. Each tree template must be specified on one line. Blank lines and lines beginning with "#" are ignored. A tree template is a parenthesized syntactic parse tree which can be used as an underspecified template tree to unify with a real syntactic parse tree. See TreeHelper.extractNode for more details.

Throws:

java.lang.Exception - if the required input property is not defined

isInitialized

public static boolean isInitialized()

Returns:

the isInitialized

setInitialized

public static void setInitialized(boolean isInitialized)

Parameters:

isInitialized - the isInitialized to set

findFocusNode

public static edu.cmu.lti.chineseNLP.util.Tree findFocusNode(edu.cmu.lti.chineseNLP.util.Tree tree)

Finds the focus, or target, word of a question using the specified tree templates and default rules that look for question words.

Using the following algorithm:

1. Look for the head word of a phrase that has a wh-word child and more than one sibling
   • look for the object of a modifying of-PP if the word is "kind" or "type"
2. Try to extract a node using the specified tree templates in the order they appear in the file
3. Using only preterminals, look for the last consecutive NN* or JJ that follows a WDT or WP
4. Using only preterminals, look for the last consecutive NN* that follows the terminals, "how many".

the focus word is returned as soon as one is found.

Parameters:

tree - The syntactic parse tree of the question

Returns:

the focus word, or null if none was found

getHeadWordOrPhrase

public static edu.cmu.lti.chineseNLP.util.Tree getHeadWordOrPhrase(edu.cmu.lti.chineseNLP.util.Tree tree)

Extracts the head word or phrase from the given Tree node, which is assumed to be an NP. Whether a phrase should be the head is determined by looking up in WordNet all the possible phrases that can be constructed from the immediate children of the input Tree and which include the right-most child.

Parameters:

tree - the Tree node from which to extract the head word or phrase

findFocusWord

public static java.lang.String findFocusWord(edu.cmu.lti.chineseNLP.util.Tree tree)

Finds the focus word, given a Tree.

Parameters:

tree - The syntactic parse tree object

Returns:

the focus word as a String or null, if one does not exist

findFocusWord

public static java.lang.String findFocusWord(java.util.List<edu.cmu.lti.javelin.qa.Term> terms)

Given a list of Terms, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.

Parameters:

terms - The list of Terms in the question.

Returns:

the focus word as a String or null, if one does not exist

findFocusWord

public static java.lang.String findFocusWord(java.lang.String question)

Given a sentence, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.

Parameters:

question - the input question

Returns:

the focus word as a String or null, if one does not exist

findFocusTerm

public static edu.cmu.lti.javelin.qa.Term findFocusTerm(java.util.List<edu.cmu.lti.javelin.qa.Term> terms)

Given a list of Terms, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.

Parameters:

terms - The list of Terms in the question.

Returns:

the focus word as a Term or null, if one does not exist

findFocusTerm

public static edu.cmu.lti.javelin.qa.Term findFocusTerm(java.lang.String question)

Given a sentence, builds a parse tree using Charniak's parser, and then uses the resulting parse tree to find the focus words.

Parameters:

question - the input question

Returns:

the focus word as a Term or null, if one does not exist

findFocusTerm

public static edu.cmu.lti.javelin.qa.Term findFocusTerm(edu.cmu.lti.chineseNLP.util.Tree tree)

Finds the focus word, given a Tree.

Parameters:

tree - The syntactic parse tree object

Returns:

the focus word as a Term or null, if one does not exist

main

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

Extracts and prints out the focus word for each question, given a file of questions.

Parameters:

args - command-line args: "<propertiesFile> <inputQuestionsFile>"

Throws:

java.lang.Exception