资源描述：
《Based on Equivalent Pseudowords基于等价伪造》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、UnsupervisedChineseWordSenseDisambiguationBasedonEquivalentPseudowordsLuZhimaoLiuTingLiShengInformationRetrievalLaboratoryofComputerScience&TechnologySchool,HarbinInstituteofTechnologyHarbin,China,150001lzm@ir.hit.edu.cnAbstract:Thispaperpresentstheconceptofequiv
2、alentpseudowordsanddevelopsanewunsupervisedChinesewordsensedisambiguationmethod.Thismethod,afterstatisticallearningfromtheequivalentpseudowords,realizesunsupervisedwordsensedisambiguationbycalculatingmutualinformationtomeasurethedegreeofcollocationinformationbetw
3、eentheambiguouswordsandtheircontext.Inourexperiment,wetesttenrealambiguouswordsandgetthehighestaccuracyof97%.Themeanaccuracywas88.52%.Theexperimentalresultshaveprovedthefeasibilityofthismethod.1IntroductionNaturelanguageisinherentlyambiguous.Awordcanhavemultiplem
4、eanings(orsenses).Givenanoccurrenceofawordwinanaturallanguagetext,taskofWordSenseDisambiguation(WSD)istodeterminethecorrectsenseofwinthatcontext.WSDisafundamentalandcentralopenproblemofNaturalLanguageProcessing(NLP).HighlyambiguouswordsposecontinuingproblemsforNL
5、Papplications.TheycanleadtoirrelevantdocumentretrievaltranslationsinMachineTranslationsystems(Palmeretal.,2000).Lexicalambiguityissyntacticorsemantic.Aword’ssyntacticambiguitycanberesolvedbyapplyingpart-of-speechtaggerswhichpredictthesyntacticcategoryofawordintex
6、tswithhighlevelsofaccuracy(Brill,1995;Brants,2000).Theproblemofresolvingsemanticambiguity,whichisgenerallyknownasWSD,hasprovedtobemoredifficultthansyntacticdisambiguation.Theonlywaytodeterminethemeaningofawordinaparticularusageistoexamineitscontext.Onecouldenvisa
7、gebuildingaWSDsystemusinghandcraftedrulesorknowledgeobtainedfromlinguists.Suchanapproachwouldbehighlylabor-intensive,withquestionablescalability.AnotherapproachinvolvestheuseofdictionaryorthesaurustoperformWSD.TherearealsothreewaystoapproachWSD:aknowledge-basedap
8、proach,whichusesanexplicitlexicon,corpus-baseddisambiguation,wheretherelevantinformationaboutwordsensesisgatheredfromtrainingonalargecorpus,or,thirdalternative
