an automatic approach finding metadiscourse mar寻找话语标志自动方法.pdfVIP

Anautomaticapproachtofindingmetadiscoursemarkers

Metadiscoursemarkersareknowntobetofspecificsemantictopics.

Hithertoresearchershavefounddozensofwaystolocatethembyfrequency,clusteror

othermethods.However,theminingformetadiscoursemarkersisnotaneasytask.In

Zhang’sstudy(2016),ahugeamountofmanualworkseemedtobeunavoidable.This

inevitabilitycouldbeaninsurmountableburdentodata-basedlinguisticstudies.Inthis

article,Iintroduceanapproachtofindingmetadiscoursemarkersautomatically,and

analyzetheefficiencyandaccuracyofit.

Keywords

Metadiscourseanalysis,metadiscoursemarkers,distributedrepresentation,machine

learning

Introduction

Metadiscoursemarkersarelinguisticexpressioncludinggrammarandpunctuation

markers.Zhang’sworkdemonstratedthesubtleinteractionbetweendifferentcategories

ofmetadiscoursemakers,andshowedthatthedistributionofacertaincategoryvaries

withthegenre.Inthisacticle,wefocusonsomeaspectsonwhichmetadiscoursecan

beseenintomorecomprehensiveandcomprehensible.

Firstly,themanualprocessduringthedataminingneedtobehandled.Forwecan

neithertrustitnorwithstandit.Thesubjectiveness,slowness,inaccuracyand

nonrepeatabilityareknowntobethefatalflawsofmanualprocess.Ontheotherhand,

weallknowthatautomaticapproachesarequick,comprehensive,comparableand

manageable.

Secondly,theclassificationinthearticleitselfisnothingbutasubjectivejudgment,

hencebringsdownthecredibilityoffurtherresearching.Byusingonlyonecorpusfor

bothmodel-trainingandmodel-testingisnotsosoundanapproachtoverifytheutility

ofmultidimensionalanalysis.

Inthisarticle,wewillmainlyexamineanalgorithmtominemetadiscourse