Data Quality from Crowdsourcing(conference)

《Data Quality from Crowdsourcing(conference)》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、NAACLHLT2009ActiveLearningforNaturalLanguageProcessing(ALNLP-09)ProceedingsoftheWorkshopJune5,2009Boulder,ColoradoProductionandManufacturingbyOmnipressInc.2600AndersonStreetMadison,WI53707USAEndorsedbythefollowingACLSpecialInterestGroups:•SIGNLL,SpecialInterestGroupforNaturalLanguageLearning

2、•SIGANN,SpecialInterestGroupforAnnotationc2009TheAssociationforComputationalLinguisticsOrdercopiesofthisandotherACLproceedingsfrom:AssociationforComputationalLinguistics(ACL)209N.EighthStreetStroudsburg,PA18360USATel:+1-570-476-8006Fax:+1-570-476-0860acl@aclweb.orgISBN978-1-932432-40-4iiIntr

3、oductionWelcometotheworkshoponActiveLearningforNaturalLanguageProcessing!Westartedorganizingthisworkshopinmid-2008afterstrongencouragementinresponsetosomeofourownworkinthearea.Aswegatheredmembersoftheprogramcommittee,thetimelinessofthetopicresonatedwithseveralofthem:thegrowingbodyofknowledge

4、onactivelearningandonactivelearningforNLPinparticularmakesthistopiconeworthexploringinafocusedworkshopratherthaninisolatedpapersinoccasional,far-flungconferences.Labeleddataisaprerequisiteformanypopularalgorithmsinnaturallanguageprocessingandmachinelearning.Whileitispossibletoobtainlargeamoun

5、tsofannotateddataforwell-studiedlanguagesinwell-studieddomainsandwell-studiedproblems,labeleddataarerarelyavailableforlesscommonlanguages,domains,orproblems.Unfortunately,obtaininghumanannotationsforlinguisticdataislabor-intensiveandtypicallythecostliestpartoftheacquisitionofanannotatedcorpu

6、s.Ithasbeenshownbeforethatactivelearningcanbeemployedtoreduceannotationcostsbutnotattheexpenseofquality.WhilediverseworkoverthepastdecadehasdemonstratedthepossibleadvantagesofactivelearningforcorpusannotationandNLPapplications,activelearningisnotwidelyusedinmanyongoingdataannotationtasks.Muc

7、hofthemachinelearningliteratureonthetopichasfocusedonactivelearningforclassificationproblemswithlessattentiondevotedtothekindsofproblemsencounteredinNLP.Relatedtopicssuchasdistributed“humancomputation”,cost-sensitivemachinelearning,andsemi-supervise