资源描述：
《The Present Use of Statistics in the Evaluation of NLP Parsers》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、
2、ThePresentUseofStatisticsintheEvaluationofNLPParsersIiJ.EntwisleD.M.W.PowersIFlindersUniversityofSouthAustraliaFlindersUniversityofSouthAustraliajim@cs,flinders,edu.aupowers~cs,flinders,edu.aumAbstracthybrids]seemcapableofexceedingthe97%barrier....Next,anewsystem..usesWearec
3、oncernedthatthequalityofresultsonlylinguisticdistributionalrules.TestedproducedbyanNLPparserbearslittle,ifagainsta38,000-wordcorpusofpreviouslyany,relationtothepercentage-resultsclaimedbythevariousNLPparser-systemspresentlyunseentext,thetaggerreachesabetterac-mavailableforuse
4、.Toillustratethisproblem,curacythanprevioussystems(over99%).weexamineonereadilyavailableNLPtagging(Voutilainen,1995)andparsingsystem,theENGCGparser;andmonetagger,theBrintagger.WenoteresponsesWeareconcernedaboutthemisleadingnatureoftobothartificiallygeneratedandnaturallyoc-suc
5、hpublishedstatistics,althoughresearcherswork-curringtext.ThepercentageassessmentsareingonNLPsystemsareofcoursewellawarethatmethodologicallyflawed,andshouldbetakenthefiguresmustbeinterpretedcarefully.Tobeablemwithagrainofsalt;instead,assessmentoftheperformanceofanNLPparsershou
6、ldbeef-toshowourcauseforconcernwemustlookatthefectedbyauser,andsolelyfromaconsidera-statisticsofawellknownparser,andalsoatitsac-tionoftheresultingparsesofexactlythein-tions.UnfortunatelytherearerelativelyfewsystemsmputwhichanNLPuserdecidestocontributewhicharefreelyavailablefo
7、rconsultationorexami-forsuchanassessment.Carefulattentiontonation,andforthisreasonweareforcedtopickoninputofwhatevercorpustheuserdecideson,ispresentlytheonlysuitablequalifyingtestofsomeofthefewsystemsthatare.Thecriticismsweparsingability.Theparsersavailablearenonemakeherearen
8、otdirectedagainsttheparsersweofthemperfectibleyet,despiteapparentyieldsuseasexamples,butagainstthewayinwhichournowquotedat99%+.Weconsidertheimpactfieldistreatingitsstatistics.ofZipf'sargumentof'leasteffort'onpercent-OurmainexampleistheENGCGparser,inoneageassessment;andweopena
9、discussiononestimatingtherelativecomplexitiesofcorpora.specificformw
