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..usesWeareconcerned
3、thatthequalityofresultsonlylinguisticdistributionalrules.TestedproducedbyanNLPparserbearslittle,ifagainsta38,000-wordcorpusofpreviouslyany,relationtothepercentage-resultsclaimedbythevariousNLPparser-systemspresentlyunseentext,thetaggerreachesabetterac-mavailableforuse.Toillustratethi
4、sproblem,curacythanprevioussystems(over99%).weexamineonereadilyavailableNLPtagging(Voutilainen,1995)andparsingsystem,theENGCGparser;andmonetagger,theBrintagger.WenoteresponsesWeareconcernedaboutthemisleadingnatureoftobothartificiallygeneratedandnaturallyoc-suchpublishedstatistics,alt
5、houghresearcherswork-curringtext.ThepercentageassessmentsareingonNLPsystemsareofcoursewellawarethatmethodologicallyflawed,andshouldbetakenthefiguresmustbeinterpretedcarefully.Tobeablemwithagrainofsalt;instead,assessmentoftheperformanceofanNLPparsershouldbeef-toshowourcauseforconcernw
6、emustlookatthefectedbyauser,andsolelyfromaconsidera-statisticsofawellknownparser,andalsoatitsac-tionoftheresultingparsesofexactlythein-tions.UnfortunatelytherearerelativelyfewsystemsmputwhichanNLPuserdecidestocontributewhicharefreelyavailableforconsultationorexami-forsuchanassessment
7、.Carefulattentiontonation,andforthisreasonweareforcedtopickoninputofwhatevercorpustheuserdecideson,ispresentlytheonlysuitablequalifyingtestofsomeofthefewsystemsthatare.Thecriticismsweparsingability.Theparsersavailablearenonemakeherearenotdirectedagainsttheparsersweofthemperfectibleye
8、t,despiteapparentyieldsuseasexamples,butagainstthewayinwhichournowquotedat99%+.Weconsidertheimpactfieldistreatingitsstatistics.ofZipf'sargumentof'leasteffort'onpercent-OurmainexampleistheENGCGparser,inoneageassessment;andweopenadiscussiononestimatingtherelativecomplexitiesofcorpora.s
9、pecificformw
