资源描述:
《Decision Trees and NLP A Case Study in POS Tagging》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、DecisionTreesandNLP:ACaseStudyinPOSTaggingOrphanosGiorgos,KallesDimitris,PapagelisThanasisandChristodoulakisDimitrisComputerEngineering&InformaticsDepartmentandComputerTechnologyInstituteUniversityofPatras26500Rion,Patras,Greece{georfan,kalles,papagel,dxri}@cti.grABSTRACTThispaperpresents
2、amachinelearningapproachtotheproblemsofpart-of-speechdisambiguationandunknownwordguessing,astheyappearinModernGreek.Bothproblemsarecastasclassificationtaskscarriedoutbydecisiontrees.Thedatamodelacquirediscapableofcapturingtheidiosyncraticbehaviorofunderlyinglinguisticphenomena.Decisiontre
3、esareinducedwiththreealgorithms;thefirsttwoproducegeneralizedtrees,whilethethirdproducesbinarytrees.Tomeettherequirementsofthelinguisticdatasets,allthreealgorithmsareabletohandleset-valuedattributes.Evaluationresultsrevealasubtledifferentiationintheperformanceofthethreealgorithms,whichach
4、ieveanaccuracyrangeof93-95%inPOSdisambiguationand82-88%inguessingthePOSofunknownwords.INTRODUCTIONIthasrecentlybecomeapparentthatempiricalMLcanfindinNLPanexcitingapplicationarea.Theincreasinguseofcorpus-basedlearninginplaceofmanualencodinghasledtotherebirthofempiricisminNLP,withprimarygoa
5、ltoovercomeaperennialproblem,namelythelinguisticknowledgeacquisitionbottleneck:foreachnew,differentorslightlydifferenttaskofNLP,linguisticknowledgebases(lexicons,rules,grammars)mostofthetimehavetobebuiltfromscratch.Anadditionalreasontopursueautomaticallyacquiredlanguagemodelsisthatitispra
6、cticallyimpossibletomanuallyencodealltheexceptionsorsub-regularitiesoccurringeveninsimplelanguageproblems,orgiveemphasistothemostfrequentregularities.Corpus-basedapproacheshavebeensuccessfulinmanyareasofNLP,butitisoftenthecasethatlanguageisbeingtreatedlikeablack-boxsystemsimulatedbylarget
7、ablesofstatistics.Although,fromtheengineeringpoint-of-view,itisawide-spreadpracticetoconsidersystemsasblackboxes,itisobviousthatthisopaquenessmakesitdifficulttounderstandandanalyzeunderlyinglinguisticphenomenaand,consequently,theimprovementofthelanguagemodelmaydepen