A Tool for Automated Revision of Grammars for NLP Systems

《A Tool for Automated Revision of Grammars for NLP Systems》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、AToolforAutomatedRevisionofGrammarsforNLPSystemsNandaKambhatlaandWlodekZadroznyIBMT.J.WatsonResearchCenter30SawMillRiverRoad,Hawthorne,NY,10532{nanda,wlodz}@us.ibm.comrulesspecifyingthesetofallowedutterancesAbstractandpossiblyalsospecifyingthesemanticsassociatedwiththeseutterances.Forinstance,

2、Wepresentanalgorithmandatoolforgrammarsarecommonlyusedinspeechautomaticallyrevisinggrammarsfornaturalunderstandingsystemstospecifyboththesetoflanguageprocessing(NLP)systemstoallowedsentencesandtospecify"tags"todisallowspecificallyidentifiedsentencesorextractsemanticentities(e.g.the"amount"ofse

3、tsofsentences.Wealsooutlineanmoney).approachforautomaticallyrevisingattributevaluegrammarsusingcounter-examples.ConstrainingthenumberofsentencesacceptedDevelopinggrammarsforNLPsystemsthatbyagrammarisessentialforreducingarebothgeneralenoughtoacceptmostmisinterpretationsofuserqueriesbyanNLPsente

4、ncesaboutadomain,butconstrainedsystem.Forinstance,forspeechunderstandingenoughtodisallowothersentencesisverysystems,ifthegrammaracceptsalargenumbertedious.Ourapproachofrevisinggrammarsofsentences,thenthelikelihoodofrecognizingautomaticallyusingcounter-examplesutteredsentencesasrandom,irrelevan

5、t,orgreatlysimplifiesthedevelopmentandundesirablesentencesisincreased.Forrevisionoftightlyconstrainedgrammars.transactionprocessingsystems,misrecognizedWehavesuccessfullyusedourtooltowordscanleadtounintendedtransactionsbeingconstrainover-generalizinggrammarsofprocessed.Aneffectiveconstrainingg

6、rammarspeechunderstandingsystemsandobtainedcanreducetransactionalerrorsbylimitingthehigherrecognitionaccuracy.numberofsentencelevelerrors.Theproblemofover-generalizationofspeechgrammarsand1IntroductionrelatedissuesiswelldiscussedbySeneff(1992).Naturallanguageprocessingsystemsoftenconstrainthes

7、etof"utterances"fromauserThus,speechgrammarsmustoftenbalancethe(spoken,typedin,etc.)tonarrowdowntheconflictingrequirementsofpossiblesyntacticandsemanticresolutionsofacceptingawidevarietyofsentencestotheutteranceandreducethenumberofincre