资源描述：
《Test Suites for Quality Evaluation of NLP Products》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、TestSuitesforQualityEvaluationofNLPProductsFrederikFouvryandLornaBalkanDepartmentofLanguageandLinguistics,UniversityofEssexWivenhoePark,ColchesterEssexCO43SQ,UK(fouvry
2、balka)@essex.ac.ukAbstractandshowshowthetestsuitecanbeusedinprac-ticetoprovideareliableassessmentmethodof
3、thelinguisticcapabilitiesofasystem.TestsuitesareausefulevaluationtoolfordevelopersandusersofNLPprod-ucts.Thepapergivesanoverviewofthe2DesignofTestSuitestsnlpdesignandmethodologyandde-scribeshowthetsnlpdataandmethod-Bytestsuiteismeantamoreorlesssystematicologycanbeusedinpra
4、cticetoprovidecollectionofspeciallyconstructedlinguisticex-areliableassessmentmethodofthelin-pressions(testitems,e.g.sentences),perhapswithguisticcapabilitiesofNLPproducts.associatedannotationsanddescriptions.Mostex-istingpubliclyavailabletestsuiteshavebeenwrit-tenforspeci
5、 csystemsorsimplycontainnumbersofinterestingexamples".Theyhaveonlylim-itedorveryspeci cannotations.Fortestsuites1Introductiontobereusable
6、sincetherearefewpubliclyavail-Testsuiteshavelongbeenrecognizedasausefulabletestsuitesaround,andhencealotofworkisevaluationtoolforNLPpr
7、oducts,especiallyforduplicatedbywritingsuitesforspeci csystem
8、thedeveloper.Untilrecently,however,therehavetheremustbeannotationstoprovideaccessviabeenfewpubliclyavailabletestsuitesorguidelinesmanydi erentsearchkeys.fortheirconstructionanduse.TheEU-fundedThisrequirementlead
9、sustowardsalarge,sys-projecttsnlp(TestSuitesforNaturalLanguagetematicallyconstructedtestsuitethatcanbe1Processing,LRE62-089),has lledthatgapbyappliedtoawidevarietyofapplications.TheprovidingamethodologyfortheconstructionofHPtestsuite(Flickingeretal.87)wentsometestsuites,as
10、ubstantialmultilingualtestsuitewaytowardsthatgoal.However,itsusabilityand(over4,500testitemsineachofitsthreelan-extendibilityisrestrictedbythelackofsupport-guages,French,GermanandEnglish),mountedingdocumentationandannotations.TheDiToonadatabase,andanumberoftoolsforusetests
11、uite(Nerbonneetal.91)pioneeredmoreex-andconstruction.Thetestdataaregeneralpur-tensiveanno
