NaturalLI Natural Logic Inference for Common Sense Reasoning

《NaturalLI Natural Logic Inference for Common Sense Reasoning 》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、NaturalLI:NaturalLogicInferenceforCommonSenseReasoningGaborAngeliChristopherD.ManningStanfordUniversityStanfordUniversityStanford,CA94305Stanford,CA94305angeli@cs.stanford.edumanning@cs.stanford.eduAbstractNocarnivoreseatanimals?fCommon-sensereasoningisimportantforwThecarnivoresNocats.

2、..AIapplications,bothinNLPandmanyeatanimalseatanimalsvisionandroboticstasks.WeproposewwNaturalLI:aNaturalLogicinferencesys-ThecatNocats...eatsanimalseatmicetemforinferringcommonsensefacts–for≡instance,thatcatshavetailsortomatoesThecatareround–fromaverylargedatabaseateananimalwofknownfa

3、cts.InadditiontobeingableThecattoprovidestrictlyvalidderivations,theateamousesystemisalsoabletoproducederivationswhichareonlylikelyvalid,accompaniedFigure1:NaturalLogicinferencecastassearch.byanassociatedconfidence.WebothThepathtotheboxedpremisethecatateamouseshowthatoursystemisabletoca

4、pturedisprovesthequerynocarnivoreseatanimals,asstrictNaturalLogicinferencesontheFra-itpassesthroughthenegationrelation(f).ThisCaStestsuite,anddemonstrateitsabilitypathisoneofmanycandidatestaken;thepremisetopredictcommonsensefactswith49%isoneofmanyknownfactsinthedatabase.Therecalland91%

5、precision.edgelabelsdenoteNaturalLogicinferencesteps.1IntroductionWeapproachthetaskofdatabasecompletion:headwayinprovidingbroad-coveragemeaninggivenadatabaseoftruefacts,wewouldliketorepresentations,itremainsappealingtousehu-predictwhetheranunseenfactistrueandshouldmanlanguageasthevesse

6、lforinference.Fur-belonginthedatabase.Thisisintuitivelycastasthermore,OpenIEandsimilarprojectshavebeenaninferenceproblemfromacollectionofcandi-verysuccessfulatcollectingdatabasesofnaturaldatepremisestothetruthofthequery.Forexam-languagesnippetsfromanever-increasingcorpusple,wewouldlike

7、toinferthatnocarnivoreseatofunstructuredtext.ThesefactorsmotivateouranimalsisfalsegivenadatabasecontainingthecatuseofNaturalLogic–aproofsystembuiltontheateamouse(seeFigure1).syntaxofhumanlanguage–forbroadcoverageTheseinferencesaredifficulttocaptureindatabasecompletion.aprincipledwaywh