Abstract. In this paper, we propose a method for labelling prepositional phrases according to two different semantic role classifications, as contained in the Penn treebank and the CoNLL 2004 Semantic Role Labelling data set. Our results illustrate the dif

smallproportionsuggeststhattheprepositionsemanticrolesweretaggedonlyincertainprototypicalsituations.Second,wewereabletoachievereasonablyhighresultsevenwhenweusedacollocationfeaturesetwithfewerthan200features.Thisfurthersuggeststhatthesemanticrolesweretaggedforonlyasmallnumberofverbsinrelatively xedsituations.Third,theprepositionSRDsystemfortheCoNLLdatasetusedaverysimilarfeaturesettothetreebanksystem,butwasnotabletoproduceanywherenearcomparableresults.SincetheCoNLLdatasetisaimedatholisticSRLacrossallargumenttypes,itincorporatesamuchlargersetofverbsandtaggingscenarios;asaresult,thesemanticrolelabellingofPPsisfarmoreheterogeneousandrealisticthanisthecaseinthetreebank.Therefore,weconcludethattheresultsofourtreebankprepositionSRDsystemarenotverymeaningfulintermsofpredictingthesuccessofthemethodatidentifyingandsemanticallylabellingPPsinopentext.

AfewinterestingfactscameoutoftheresultsovertheCoNLLdataset.ThemostimportantoneisthatbyusinganindependentprepositionSRLsystem,theresultsofageneralverbSRLsystemcanbesigni cantlyboosted.Thisisevidentbecausewhentheoracledresultsofallthreesubtaskswereused,themergedresultswerearound10%higherthanthosefortheoriginalsystems,inallthreecases.Unfortunately,itwasalsoevidentfromtheresultsthatwewerenotsuccessfulinautomatingprepositionSRL.DuetothestrictnessoftheCoNLLevaluation,itwasnotalwayspossibletoachieveabetteroverallperformancebyimprovingjustoneofthethreesubsystems.Forexample,insomecases,worseresultswereachievedbyusingtheoracledresultsforVA,andtheresultsproducedbySRDclassi erthanusingtheVAclassi erandtheSRDclassi ersinconjunction.Thereasonfortheworseresultsisthatinourexperiments,theoracledVAalwaysidenti esmoreprepositionsattachedtoverbsthantheVAclassi er,thereforemoreprepositionswillbegivensemanticrolesbytheSRDclassi er.However,sincetheperformanceoftheSRDclassi erisnothigh,andthesegmentationsubsystemdoesnotalwaysproducethesamesemanticroleboundariesastheCoNLLdataset,mostoftheseadditionalprepositionswouldeitherbegivenawrongsemanticroleorwrongphrasalextent(orboth),therebycausingtheoverallperformancetofall.

Finally,itisevidentthatthemergingstrategyalsoplaysanimportantroleindeterminingtheperformanceofthemergedprepositionSRLandverbSRLsystems:whentheperformanceoftheprepositionSRLsystemishigh,amorepreposition-orientedmergingschemewouldproducebetteroverallresults,andviceversa.

5ConclusionandFutureWork

Inthispaper,wehaveproposedamethodforlabellingprepositionsemanticsanddeployedthemethodovertwodi erentdatasetsinvolvingprepositionsemantics.Wehaveshownthatprepositionsemanticsisnotatrivialproblemingeneral,andalsothathasthepotentialtocomplementothersemanticanalysistasks,suchassemanticrolelabelling.

OuranalysisoftheresultsoftheprepositionSRLsystemshowsthatsig-ni cantimprovementinallthreestagesofprepositionsemanticrolelabelling—namelyverbattachment,prepositionsemanticroledisambiguationandargu-