Walter Daelemans

《Walter Daelemans》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、WalterDaelemansUniversityofAntwerpLinguistics–CNTSUniversiteitsplein1B-2610WilrijkBelgium90.MachineLearningofNaturalLanguage1.IntroductionInthisarticleweprovideanoverviewofrecentresearchontheapplicationofsymbolicMachineLearningtechniquestolanguagedata(MachineLearningofNaturalLanguage,MLNL).

2、BothinQuantitativeLinguistics(QL)andinMLNL,themaingoalistodescribethelanguageasitisobservedwithrules,languagemodels,orotherdescriptions.ButwhereasthemotivationinQLispurelyscientific(establishingthelawsandmathematicalpropertiesoflanguage),themotivationinMLNLispartlypragmatic:increasingtheacc

3、uracyandefficiencyofNaturalLanguageProcessing(NLP)systems,orthespeedwithwhichtheycanbebuilt.Nevertheless,theextractedgeneralizationscanprovideworthwhileinsightintothelanguagetaskbeingstudiedinanMLNLframework.Therearedifferentwaysinwhichthealgorithmicacquisitionoflanguageknowledgeandbehavior

4、isstudied.Wecannotpossiblydiscussallthisrelevantworkinthecontextofthisarticle.Oneimportantareaofresearchthatwillbeomittedinthisoverviewisthecomputationalmodelingofhumanlanguageacquisitionusingstatistical,machinelearningorneuralnetworkmethods(McClellandandRumelhart1986;Brent1996;BroederandMu

5、rre2000).Salientmodelingresultsinthisareaincludethelearningofwordmeaning(Siskind1996;Resnik1996),findingstructureinsequences(Elman1990),bootstrappingwordsegmentation(BrentandCartwright1996),andsettingparametersinaprinciplesandparametersapproach(NiyogiandBerwick1994;DresherandKaye1990)aswell

6、asempiricistalternativestothelatterapproach(MacWhinneyandLeinbach1991;Gillisetal.1993;Gillis,DurieuxandDaelemans1995).Thedevelopmentofalgorithmsfortheinferenceofformalgrammarsfrom(mostlyartificial)languagedataisanotherrelevantareaofresearchthatwillnotbecoveredinthisarticle.SeeHonavarandSlut

7、zki(1998)foracollectionofresearchinthisarea.WerestrictouroverviewheretolanguagelearninginthecontextofNaturalLanguageProcessingandComputationalLinguistics.Wefirstprovideabriefintroductiontothe‘empiricalrevolution’inNLP,andtheincreasedattentionforMachineLe