语言学-语态(英文) 23页

1DitransitiveConstructions:TowardsanewRoleandReferenceGrammaraccount?MARTINHASPELMATHMax-Planck-InstitutfürevolutionäreAnthropologie,LeipzigInthispaper,IwillexaminethetreatmentofditransitiveconstructionsinRoleandReferenceGrammarandcompareittotheperspectiveonditransitivesthatIhavedevelopedinearlierwork(Haspelmath2005a,2007a),aswellastosomeotherformalgrammaticalframeworks.IwillconcludebyproposingafairlyradicalrevisionofthestandardtreatmentofditransitiveconstructionsinRRG(Guerrero&VanValin2004,VanValin2007).1.SomefoundationalprinciplesforsyntactictheorizingIhavelongbeenasympathizerofRoleandReferenceGrammar(RRG),inparticularbecauseitadoptsthreefoundationalprinciplesofsyntactictheorizingthatIregardasextremelyimportantandthathavebeenneglectedinmostotherformalsyntacticapproaches:(i)Non-apriorism:Thisprinciplesaysthatdescriptiveconceptsshouldnotbeselectedapriori,butshouldbedeterminedseparatelyforeachlanguageonthebasisoflanguage-internalevidence(cf.Croft2001,Haspelmath2007b,Frajzyngier2006).ThisprincipleisnotveryprominentinwritingsonRRG,butVanValin(2006)statedveryclearlyatthebeginningofaplenaryconferencepresentation:"RRGisanon-aprioristictheory".(ii)Typologicaladequacy:ThisprinciplesaysthatthetheoreticalapproachshouldbeapplicabletoanylanguageandnotbebiasedtowardindividuallanguagessuchasEnglish,German,LezgianorLakhota.ThisprinciplehasbeenprominentinRRGsinceitsinceptionandisperhapsthemainreasonwhyRRGlookssodifferentfromothermainstreamformaltheoriessuchasMinimalism,LFGorHPSG.(iii)Semantic-pragmaticmotivation:ItisrecognizedinRRGthatsyntaxcannotbeunderstoodseparatelyfromsemanticsandpragmatics,becausetoaverylargeextentitissemanticallyand/orpragmaticallymotivated.However,therearealsotwoprinciplesofRRGthatareproblematicandthatIwillargueshouldbereconsidered:(iv)Nonautonomyofsyntax:Syntaxisnotautonomous,andsemanticandsyntacticstatementscanbemixedfreely.Theprinciplehasnotbeenformulatedinthisway,butaswewillsee,RRGpracticesometimesshowsthataversionofitisassumed.(v)Descriptivesimplicity:Anoptimal(semanto-)syntacticframeworkshouldallowlinguiststoformulateverysimplerules.Again,IamnotawareofanexplicitstatementofthisprincipleintheRRGliterature,butitisclearfrommanyoftheargumentsforparticularanalysesthatsuchaprincipleis1oftenassumed,asitisingenerativelinguistics.Finally,Iwanttoargueagainstaprinciplethatwouldprobablynotbedefendedbyanyone:1Inapersonalcommunication,RobertVanValinconfirmsthis:"RRGhasalwaysstrivenforthesimplest,mostgeneralanalysespossiblewithaslittlestipulationaspossible."\n2(vi)PassivePrivilege:Passivizationismoreimportantthanotherbehaviouralpropertiesfortheorganizationofsyntax.Wewillseebelowthatapparentlysomethinglikethishasbeenassumed.2.DitransitiveconstructionsBeforegoingontodiscusshowvariousgrammaticalframeworksdealwithditransitiveconstructions,Iintroducehereafewconceptsthatarecrucialtomyownunderstandingoftherelevantrangeofphenomena.2.1.DelimitationDitransitiveconstructionsaredefinedhereasthree-argumentconstructionswithanactor,atheme,anda"proto-recipient"(Primus1999),i.e.anargumentthatissimilartoarecipientandoccursinathree-placeconstruction.Proto-recipientsincludethefollowingmorespecificroles:recipient-possessor(with'give','hand','donate'),goal-possessor(with'bring','send'),addressee-listener(with'tell','explain'),andaddressee-viewer(with'show'),aswellasothercloselyrelatedroles.Notincludedinthecategoryofditransitiveconstructionsareactor-theme-locationpatternsofthe'load/spray'type.Thus,followingcommonusageintheliterature,ditransitiveisnarrowerthanthree-place.Thephenomenathatareofinterestherearevariouscodingpropertiesandbehavioralpropertiesofthethemeargument(T)andtherecipientargument(R),aswellasalternationssuchastheDativeAlternationinEnglish,illustratedin(1a-b).(1)a.PrepositionalDativeConstruction:PedrogaveAisha(R)hise-mailaddress(T).b.Double-ObjectConstruction:Pedrogavehise-mailaddress(T)toAisha(R).2.2.ThethreemajoralignmenttypesIstartfromtheobservationthatinstudyingditransitiveconstructionsacrosslanguages,itishelpfultodistinguishthreemainalignmenttypes,analogoustothethreewell-knownmonotransitivealignmenttypes(cf.Haspelmath2005a,Siewierska2004:57-63).Thepicturethatisshownin(2)hasbecomestandardtextbookwisdomfortwo-placetransitiveconstructions.Ifweusethewell-knownhyperrolesS(singleargumentofintransitiveverb),A(actor,agent-likeargumentoftransitiveverb)andU(undergoer,patient-likeargumentoftransitiveverb),wecansaythatifSandAaretreatedalikeasopposedtoU,wegetaccusativealignment(asin2a);ifallthreearetreatedalike,wegetneutralalignment(asin2b);andifSandUaretreatedalikeasopposedtoA,wegetergativealignment(asin2c).\n3(2)Themajormonotransitivealignmenttypesa.Snominativeb.Sc.SabsolutiveAUaccusativeAUergativeAUaccusativealignmentneutralalignmentergativealignmentNowasBlansitt(1984)andDryer(1986)firstpointedout(seealsoCroft(1990:100-108),Dryer2007),therelationshipbetweenthetwoobjectargumentsinditransitiveclausescanbeconceptualizedinexactlythesameway.ThehyperrolesinditransitiveclausesareR(recipient-likeargument,orproto-recipient)andT(theme-likeargument).DependingonwhetheritisTorRthatistreatedlikethemonotransitiveU,wegettwodifferentnon-neutralalignmentpatternsandaneutralpattern,shownin(3a-c).InDryer's(1986)terminology,whenTistreatedlikethemonotransitiveU,wehaveadirect-object/indirect-objectdistinction.Renamingittodirective/indirective,asin(3a),makestheparalleltomonotransitivealignmentevenclearer.AndwhenRistreatedlikethemonotransitiveU,wehaveaprimary-object/secondary-objectdistinction.Again,forterminologicalconveniencethishasbeenrenamedtoprimative/secundativein(3c).(3)Themajorditransitivealignmenttypesa.Udirectiveb.Uc.UprimativeTRindirectiveTRsecundativeTRindirectivealignmentneutralalignmentsecundative2alignmentThesealignmenttypesarerelevantforanymorphosyntacticpatternthatcoulddistinguishthearguments,butforexpositoryconveniencethefollowingexamplesjustshowthesealignmenttypesasexpressedinflagging,i.e.caseandadpositionalmarking.(i)indirectivealignment:TandUshowaccusativecase(4)Germana.SanktGeorg(A)tötetedenDrachenACC(U).'St.Georgekilledthedragon.'b.SanktMartin(A)gabdemBettlerDAT(R)seinenMantelACC(T).'St.Martingavethebeggarhiscloak.'(ii)neutralalignment:U,TandRshowaccusativecase2Notethatthealignmenttypeiscalledsecundative,afterthesecondaryobject(notprimative),becausethesecundativeargumentisthespecialcase(thisisanalogoustothetermsaccusativealignmentandergativealignment).SeealsoVanValin(2005:127),whereitisnotedthattheterm"primaryobjectlanguage"createsproblems.\n4(5)Martuthunira(Pama-Nyungan;Dench1995:156,67)a.Ngayunhawu-lhakayarra-atharratal-yu(U).I.NOMsee-PSTtwobird(sp.)-ACC'Isawtwotharratalbirds.'b.Ngunhukanyarangurnujinkarn-ku(T)yungku-lhathat.NOMmanthat.ACCstick-ACCgive-PSTngurnula-ngu-umimi-i(R).that.DEF-GEN-ACCuncle-ACC'Thatmangavehisunclethediggingstick.'(iii)secundativealignment:onlyTshowsapreposition(6)Yoruba(Benue-Congo;Rowlands1969:21)a.ópamí(U)hekillme'Hekilledme.'b.ófúnmi(R)l'ówó(T)hegivemeSECmoney'Hegavememoney.'3.SomeotherapproachestoditransitiveconstructionsIwillnowlookatafewformalsyntacticframeworksandtheirwayofdealingwithditransitiveconstructions,asapreludetomydiscussionofRoleandReferenceGrammar'sapproachinthesubsequentsections.3.1.RelationalGrammarRelationalGrammar(Perlmutter1980,Blake1990),anapproachthatisnolongerwidelypracticedbutthatwashighlyinfluentialinthe1970sand1980s,wasespeciallyconcernedwithalternationssuchastheEnglishDativeAlternation.Thiswasanalyzedinawayanalogoustothepassivealternation.RelationalGrammarasssumedthethreecoregrammaticalrelations"1"(subject),"2"(directobject),and"3"(indirectobject),asshownin(7)and(8)belowthecorearguments.(7)Pedrogavehise-mailaddresstoAisha.123(8)PedrogaveAishahise-mailaddress.132(initial)12Chômeur(final)The"dative-shifted"formofthealternationin(8)isdescribedbytheoperationof3-to-2Advancement,whichchangesthegrammaticalrelationoftherecipientAishafrom3attheunderlying("initial")stratumto2atthesurface("final")stratum.Sinceeachgrammaticalrelationcanoccuronlyonceperclause,theformer2(thethemehise-mailaddress)isturnedintoachômeur,aspecialgrammaticalrelationforanounphrasewithanunderlyingcorerelationthatwasoustedbyanadvancementprocess.Theparallelwiththe\n5passiveconstructionsshownin(9)-(10):Passivizationisanalyzedas2-to-1Advancement,andtheagentphraseisachômeur.(9)AishacriticizedPedro.12(10)PedrowascriticizedbyAisha.21(initial)1Chômeur(final)Problemswiththisapproachbecameapparentsoon.Oneproblemisthatthesystemconsistingoftheuniversalcorerelations1,2,3andallowingnodoublingofgrammaticalrelationscannothandlecaseswithneutralalignment,wheretherecipientandthethemeargumentshavethesamesyntacticproperties.Gary&Keenan(1977)citedtheexampleoftheBantulanguageKinyarwanda,whichtheyclaimedmustbeanalyzedashavingtwo(direct)objects.FortheparticularcaseofKinyarwanda,Dryer(1983)claimedthattherearesomeminordifferencesbetweentherecipientandthethemeobjectafterall,sothattheRelationalGrammaraccountwouldstillbeviable.However,itisunclearhoweasilythissolutioncanbegeneralized.Anotherproblem,highlightedbyDryer(1986)isthefactthatcasesofsecundativealignmentcanbehandledonlybyobligatory3-2Advancement,asproposed,forinstance,byAissen(1983,1987)forTzotzil,aMayanlanguageofMexico.Thisisasifoneweretohandleergativitybyobligatorypassivization,anapproachthatwaswidespreadinthe19thcentury,butdoesnotseemacceptableanymore.Toaddressthisissue,Dryer(1986)introducedtheprimary/secondaryobjectdistinction,butwithoutabandoningtheprimacyofRelationalGrammar's2/3contrast.Itwasonlyinlaterwork,startingwithCroft(1990:100-108),thatthesecondary/primarycontrastwasseenasfullyparalleltoandonanequalfootingwiththeindirect/directcontrast.3.2.FunctionalGrammarFunctionalGrammar(Dik1989,1997)isamonostrataltheorythatisclosertoRoleandReferenceGrammarinitsbasicarchitecture,butitsSemanticFunctionscorrespondfairlycloselytoRelationalGrammar'sinitialstratum,whileFG'sSyntacticFunctionscorrespondtoRelationalGrammar'sfinalstratum.EachargumentisassignedaSemanticFunction,andanargumentmayadditionallybeasssignedoneofthetwoSyntacticFunctionsSubjectandObject.CorrespondingtoRelationalGrammar'sinitial1,2and3,wehaveAgent,PatientandRecipient,andcorrespondingtoRelationalGrammar'sfinal1and2,wehaveSubjectandObject.Correspondingtothechômeur,FunctionalGrammarhasAgentandPatientargumentswithoutaSyntacticFunction.Thisisillustratedforpassivizationin(11)-(12),andfortheDativeAlternationin(13)-(14).(11)AishacriticizedPedro.AgPatSubjObj\n6(12)PedrowascriticizedbyAisha.PatAgSubj(13)Pedrogavehise-mailaddresstoAisha.AgPatRecSubjObj(14)PedrogaveAishahise-mailaddress.AgRecPatSubjObjThissystemavoidsthechoicebetween(13)and(14)astheunderlyingstructurethatwasaproblemforRelationalGrammar(asalsodiscussedbyDryer1986),becauseneitherismore"basic"thantheother.ButFunctionalGrammarhasthesameproblemsasRelationalGrammarwithneutralditransitivealignmentasinKinyarwanda,anditalsohasthesameproblemswithsecundativealignment.Interestingly,thesolutionsofferedwerequiteparallel:WhereRelationalGrammarhadtoassumeobligatory3-to-2Advancement,Dik(1989:240-241;1997:282-289)hadtoassumeFG'scounterpartofthis,obligatoryObjectassignment.ThisiscompletelyagainstthespiritofDik'sSyntacticFunctions,whicharesupposedtoberelevantonlywhenanalternationexists(i.e.Subjectassignmentissupposedtoberestrictedtolanguageswithpassivization,andObjectassignmentissupposedtobe3restrictedtolanguageswithadativealternation;seealsoSiewierska1998).3.3.LexicalDecompositionGrammarLexicalDecompositionGrammar(Wunderlich1997,2006)isanotherambitiousattemptataccountingforthepropertiesofverbalargumentsinasystematicwayfromacross-linguisticperspective.LDGoperateswithgrammatical-relationsfeaturesonarguments,asillustratedintheargumentstructureofGermantöten'kill'in(15).Thisconsistsofthedecompositionontherighthandside('xactsandtherebyycomestobedead'),andthetwoargumentsmarkedbylambdas,annotatedbythefeatures[±hr]("thereisahigherrole/thereisnohigherrole")and[±lr]("thereisalowerrole/thereisnolowerrole").Thefeature[+hr]correspondsroughlytoRoleandReferenceGrammar'sundergoer,and[+lr]correspondstoactor(Wunderlich2006:65).(15)tötenλyλx{ACT(x)&BECOMEDEAD(y)}+hr–hr–lr+lr(dir.object)(subject)InLDG,casesalsohavefeaturesofthesametype,andtheymustmatchthefeaturesofthearguments.Nominative/absolutivecasehasthefeaturespecification[](i.e.completeunderspecification),accusativeis[+hr],andergativeis[+lr].Ifalanguagehasanaccusativeandanominative/absolutive,accusativemustgoonthedirectobjectandnominative/absolutiveonthe3FunctionalGrammar,likeRelationalGrammar,alsohasaproblemwithergativealignment,andinterestingly,Dik(1997:284-289)suggestsasimilarsolutiontoergativity:Itcouldbeexplained(diachronically)byobligatorypassivization.\n7subject,becauseotherwisethefeaturesdonotmatch.If,however,alanguagehasanergativeandanominative/absolutive,ergativemustgoonthesubjectandnominative/absolutiveonthesubject.Thus,likeRRG'sActor/Undergoersystem,LDG'sfeaturesystemisdesignedtoaccountbothforaccusativeandforergativealignment.Inthisregard,LDGandRRGareclearlysuperiortoRelationalGrammarandFunctionalGrammar.Nowwhathappensinditransitiveconstructions?ConsiderGermangeben'give'in(16).Herethedecompositionis'xactsandtherebyycomestopossessz'),andthethreeargumentsmarkedbylambdasareshownin(16)aswell(Wundeerlich2006:113):(16)geben:λzλyλx{ACT(x)&BECOMEPOSS(y,z)}+hr+hr–hr–lr+lr+lr(dir.object)(ind.object)(subject)Theindirectobjectisbetweenthesubjectandthedirectobjectinthehierarchyofroles,soitgetsboththefeature[+hr](becausethesubjecthasahigherrole)and[+lr](becausethedirectobjecthasalowerrole).Ifitisassumedthatthedativecasehasthefeaturespecification[+hr,+lr](moremarked,correspondingtothefactthattypicallythedativecaseismorphologicallymoremarkedthantheaccusativeorergative),thelinkingbetweenthecasesandtheargumentsworks:Whilebothnominativeandaccusativewouldbeabletounifywiththeindirectobject's[+hr,+lr],the4dativetakesprecedencebecauseitisthemorespecificcase.LDGhasthesameproblemsasRelationalGrammarandFGwithneutralalignment,butWunderlichdoesnotdiscusssuchcases.However,hedoesdiscusssecundativealignment,andunfortunately,hiselegantsystembecomesmuchlessneathere.In(17),weseeWunderlich'slexicalentryforaverbsuchasYorubafún'give'.Thefeature[±hr],whichdistinguishesbetweenthesubjectandtheobjects,isdistributedinthesameway,butinsteadofthefeature[±lr],anewfeature[±ho]("thereisahigherobject/thereisnohigherobject")isusedtodistinguishbetweenthetwoobjects.(17)fún:λzλyλx{ACT(x)&BECOMEPOSS(y,z)}+hr+hr–hr+ho–ho(sec.object)(prim.object)subjectThisisinelegant,becausethesecundativeconstructionneedsnewmachinery,whiletheindirectiveconstructioncanbedescribedwiththesamemachinerythatisindependentlyneededtoaccountfortheaccusative/ergativecontrast.Moreover,thesystemcannothandlelanguageswithergativemonotransitivealignmentandsecundativeditransitivealignment(Wunderlich2006:137).ApparentlythefactthatEuropeanlanguagesoverwhelminglyshowthe4Interestingly,LDG'streatmentofthedativeistheexactoppositeofRRG's,wherethedativeisconsideredthedefaultcase(VanValin2005:110).ForLDG,animportantfurtherconsiderationisthatthedativealsotendstohavethemostspecificovertmarking,comparedtonominative/absolutive(whicharegenerallyzero-coded)andaccusative/ergative(whichareatleastsometimeszero-coded),sothatovertcoding("formalmarkedness")correspondstoadoublypositivefeaturespecification("functionalmarkedness").(ThisisofnoconcerntoRRG,whicharguesthatcauseemarkingbydativecasesupportstheviewofthedativeasthedefaultcase;VanValin&LaPolla1997:§9.2.2.)\n8indirectiveconstructionhasinfluencedthedesignofthetheory.4.RoleandReferenceGrammarReadersofthisvolumewillbyandlargebefamiliarwiththemainfeaturesofRoleandReferenceGrammar,sothisoverviewcanbebrief.4.1.ArgumentstructuresInRRG,theargumentstructuresaredeterminedbypossibleLogicalStructures,whichareseenaspropertiesofthehumanconceptualsystem.Table1showsthewell-knownlistofpossibleLogicalStructuresaccordingtotheRRGsystem,witheachLogicalStructurecorrespondingtoabasicverbclass.Table1:LexicalrepresentationforAktionsartclasses(fromVanValin2007)Allditransitivepredicates(inthesenseof§2.1above)areregardedashavingaLogicalStructureofthetypein(18),whereanagentxactstocausearecipientytobeinsomepredicaterelationtothethemez.(18)[do´(x,Ø)]CAUSE[BECOMEpredicate´(y,z)]Inthecaseoftransferverbs,thepredicateishave´,andforthesakeofsimplicity,wecanlimitourselvestotransferverbsinthisarticle.Formentalditransitiveverbslike'show',wherethepredicatewouldnotbehave´,theanalysiswouldbebasicallythesame.4.2.CodingandalternationsInEnglish,therearetwopossibilitiesforrealizingthislogicalStructure.Moststraightforwardly,thethemecanbecomeUndergoer,followingtheuniversalActor-UndergoerHierarchyinFigure1.Accordingtothishierarchy,thetheme(thesecondargument(z)ofthepredicate´in(18))is"lessmarked"(i.e.universallypreferred)asUndergoerthantherecipient(thefirstargumentofthepredicate´in(18)).\n9Figure1:TheActor-UndergoerHierarchyThis"default"choiceofUndergoeryieldstheEnglishPrepositionalDativeConstruction(PatgavethebooktoKim).Theconstituentstructure,theLogicalStructureandthelinkingtoActorandUndergoerareshowninFigure2(fromVanValin2007).ThechoiceoftheprepositiontofortherecipientalsofollowsafairlygeneralruleforEnglishto(discussedinVanValin&LaPolla1997:376-377,VanValin2005:113).Figure2:TheEnglishPrepositionalDativeConstructionButEnglishallowsasecondpossibility,theDouble-ObjectConstruction(PatgaveKimthebook),wheretherecipientischosenasundergoer.ThislinkingisshowninFigure3(againfromVanValin2007).Figure3:TheEnglishDouble-ObjectConstructionThischoiceofundergoerisnotderivedfromgeneralprinciples,butisspecificallystipulatedforthisconstruction.Itisalsocalled"markedundergoerselection"(VanValin2005:61,2007:§3).\n10Thus,whattheclassicalRoleandReferenceGrammardescriptionshareswithRelationalGrammar,FunctionalGrammarandLexicalDecompositionGrammaristhattheindirectivepatternisprivilegedoverthesecundativepattern.ThenextsectionwilldiscusstheRRGdescriptionandsomeofitsproblemsingreaterdetail.5.ProblemswiththeRRGanalysis5.1."Marked"UndergoerselectionWhyshouldoneofthepossibilitiesforcodingtheditransitiveconstruction,namelytheindirectivelyalignedpattern,beconsidered"unmarked"ordefault,whiletheotheroneis"marked"?Theterm"(un)marked"hasmanydifferentsenses(cf.Haspelmath2006),sothefirstquestionisinwhatsensethechoiceofundergoeris"unmarked".Anobviouspossibilityis"unmarked"inthesenseof"normal",orcross-linguisticallyfrequent.Itistruethattheavailablecross-linguisticevidenceindicatesthattheindirectivepatternismorefrequentthanthesecundativepatternandtheneutralpattern.However,itisnotmuchmorefrequent:Haspelmath(2005b)providesthefiguresinTable2,basedonaworld-widesampleof339languages.languagesgenerafamiliesIndirect-objectconstruction18910453Double-objectconstruction845120Secondary-objectconstruction665130Table2:Thedominantditransitivealignmentpatternsin339languages(basedonHaspelmath2005b)Alsowithinlanguagesthathavebothpatterns(about10%ofthelanguages,bothinSiewierska's(1998)dataandinHaspelmath's(2005a)data),itisnotthecasethatthedativepatternisnecessarilythemorefrequentpattern.InEnglish,forexample,thedouble-objectpatternisclearlymorefrequent,atleastinthespokenlanguage.VanValin(2005:62)givesseveralfurtherreasonsforconsideringthePrepositionalDativeConstructioninEnglish"unmarked".OneisthattheDouble-ObjectConstructionismoreconstrainedanddoesnot,forexample,allowpersonalpronounsinthemeposition(*MarygaveJohnthem).ButthiscanalsofollowfromthefactthatthePrepositionalConstructionismoreexplicitlycoded.Thatconstructionswithclearercodinghavegreaterpossibilitiesofcombinationisnormal(as,e.g.,inthecaseofcomplementclauseswithandwithoutthat)andperfectlyunderstandablefromafunctionalpointofview(cf.Rohdenburg1996);noappealto"markedness"isneeded.Anotherreasonisthatinsyntheticcompoundslikeflowergiverandgirlgiver,thefirstcompoundmemberisinvariablyinterpretedasthetheme.Butthisisevidentlycloselyrelatedtothefactthatinsuchquasi-incorporationstructures,theargumentisnon-referential,andhumanrecipientsare\n11normallyreferential.Eveninlanguagesinwhichtherecipientisalwaysthe5undergoer,onlythemescannormallybeincorporated.Thus,onewonderswhethertheunequaltreatmentoftheindirectivepatternandthesecundativepatterninRRGisafeaturethatwasinheritedfromtransformationalapproaches,whereoneofthealternatingpatternsisregardedas"underlying/initial",whilethetheotheroneis"derived/final".RRGisamonostrataltheory,sotherewouldbenointrinsicreasontogiveprivilegedtreatmenttoonepatterninanalternation,andthealternativeRRGaccountsuggestedbelowin§6treatsthembothequally.Butevenifoneofthetwopatterns(indirectiveandsecundative)istobegivenaprivilegedtreatment,onewouldstillhavetoshowthatthesecundativepatterncouldnotbetheprivileged,morebasicpattern.Inderivationalaccounts,ithasoccasionallybeenproposedthattheindirectivepatternshouldbederivedfromthesecundativeorthedouble-objectpattern.Mostprominently,Dryer(1986)proposedan"AntidativeShift"forEnglishintheRelationalGrammarframework.Accordingtothisapproach,thedouble-objectconstructionin(19)isprimary,andthedativeconstructionin(20)isderivedfromanunderlyingstructurelike(19)bySO-to-PO(secondarytoprimaryobject)advancement.(19)PedrogaveAishahise-mailaddress.SUPOSO(20)Pedrogavehise-mailaddresstoAisha.SUSOPOSUPOChômeurInasimilarway,RRGcouldmodifytheActor-UndergoerHierarchyinFigure1insuchawaythatthesecundativepatternbecomesthe"unmarked"choice,butthiswouldprobablyhaveundesiredrepercussionselsewhere.Overall,onegetstheimpressionthatanimportantreason(perhapsthemostimportantreason)forsayingthatthethemeundergoeris"unmarked"isthatitfitsbetterintotheoverallRRGsystemandmakesitsimpler(cf.theunstatedprincipleofdescriptivesimplicitymentionedin§1).Thisisagainthatcomesatthepriceofrelativeunmotivatednessofthechoiceofwhichargumentisthe"unmarked"undergoer.InGuerrero&VanValin(2004)andVanValin(2007),arevisedRRGaccountispresented,whichrecognizesthat"themarkednessrelationsexpressedinFigure1arenotvaliduniversally".Thisaccountthusabandonstheideaofauniversaldefault,althoughtheideathattheremaybealanguage-particulardefaultisstillretained.Iwilldiscussthisrevisedaccountbelowin§7.AlthoughVanValin(2005:123-127)adoptsthisrevisedaccount,hestillpresentsargumentsforthemarkednessapproachelsewhereinthebook(p.61-62),whichiswhyIhaveincludedthisdiscussionhere.5VanValin(2005:62)alsoobservesthat"dative-shift"alternationsgenerallyrequireovertapplicativemarkingfortheconstructioninwhichthenon-themeistheundergoer,whichwouldagainsupportthe"markedness"ofnon-themeundergoerassignment.However,itseemsthatapplicativemarkingwithtypicalditransitiveverbsisquiterare.Mostapplicativesarebeneficiary,comitativeandinstrumentalapplicatives(Peterson2007).\n125.2.Undergoerasa"semanticmacrorole"VanValin(2004:74-78)emphasizesthatActorandUndergoeraresemanticmacroroles,i.e.thatincontrasttopurelysyntacticgrammaticalrelationsofotherframeworks,theseconceptshavesemanticimport.Giventhis,itisexpectedthatthereshouldbeasemanticdifferencebetweenatheme-undergoerandarecipient-undergoerconstruction.Thatthisissometimesindeedthecasecanbeseenclearlyinthelocativealternation(VanValin&LaPolla1997:145,VanValin2007:§3):(21)a.Oxfamloadedtheplane(U)withreliefgoods.b.Oxfamloadedreliefgoods(U)ontheplane.Itiswell-knownthat(21a)and(21b)arenotidenticalsemantically,anditisreasonabletorelatetheirdifferencestothefactthatthelocationisundergoerin(21a),whilethethemeisundergoerin(21b).Suchasemanticdifferencecanalsobeobservedfordirectional-motionverbslike'throw':(22)a.Pedrothrewtheball(U)toAisha.('inthedirectionof')b.PedrothrewAisha(U)theball.('intothepossessionof')ItisonlyinthelattercasethatitcanbeinferredthatAishacametohavetheball.Intherecentliteratureonthesepatterns,twodifferenteventstructureshaveoftenbeenposited(e.g.Goldberg1992,1995,Harley2002,Wunderlich2006:§6.6):(23)a.[do´(Pedro,Ø)]CAUSE[BECOMEbe-at´(ball,Aisha)]b.[do´(Pedro,Ø)]CAUSE[BECOMEhave´(Aisha,ball)]Giventheseeventstructures,thesemanticdifferencebetween(22a)and(22b)iseasilyexplained.(Differenteventstructuresarealsopossibleforthelocativealternationin(21),cf.Kailuweit2005).However,suchadual-event-structureapproachisnotmotivatedforallditransitiveverbsinEnglish,aswaspointedoutbyRappaportHovav&Levin(2006).Simpletransferverbslike'give','lend','show'donothavedifferentmeaningsinthetwoconstructions,andfortheseverbsthereisnoevidencethattwoeventstructuresareinvolved.Moreover,thenon-defaultchoiceofundergoermustbelexicallyspecifiedandcannotbederivedfullyfromaverb'smeaning.Inthissense,undergoerisclearlyasyntacticdevice.VanValin(2004:77-78)notesthatnondefaultmacroroleselectionisnotcompletelyarbitrary,butseemstobesemanticallymotivated–butthisisquitetypicalofsyntacticrules:Theycannotbestatedinpurelysemanticterms,buttheyarenotentirelyarbitrarysemantically.Crucially,undergoerselectionmustmakereferencetonon-semantic6information,unlike(Jackendovian)thematicrolesor(Dowtyan)proto-roles.6Thefeatures+lrand+hrofLexicalDecompositionGrammar,whichcorrespondtoactorandundergoer(aswesawin§3.3),aresimilarinthattheymayalsobeintroducedby"exceptionallexicalmarking"(Wunderlich2006:106),notjustderivedbyrulefromthedecompositionalstructure.\n13VanValin(2004:75)suspectsthatapproacheslikeManning's(1996),wherethecounterpartsofactorandundergoerhavesyntacticstatus,aremotivatedbytheassumptionoftheautonomyofsyntax.But"autonomyofsyntax"canmeantwothings:(i)syntaxshouldbedescribedandunderstoodwithoutregardforsemantics(rejectedbymostlinguists,especiallyfunctionalists)(formalistautonomy)(ii)semanticandsyntacticstatementsshouldbecarefullydistinguished(assumedbymostlinguists)(descriptiveautonomy).Itseemstomethatthereareverygoodreasonsforrejectingformalistautonomyin(i),andnogoodreasonsforrejectingdescriptiveautonomyin(ii).Infact,theformalist/functionalistdivideinlinguisticscannotbeusefullylinkedtotheautonomynotion,despitewhatsomefunctionalistsandsomeformalistshaveclaimed(especiallyCroft1995andNewmeyer1998;seeHaspelmath2000).Manning'sreasonsforclaimingthatactorandundergoerhavesyntacticstatusisprobablyjustmotivatedbydescriptiveautonomy.IwouldurgethatRRG,too,shouldadopttheprincipleofdescriptiveautonomy(cf.§1),andeitheracceptthatactorandundergoerarenot(entirely)semanticconcepts,orredefinetheirroleinsuchawaythatnonsemanticrulessuchaslexicallyspecifiedundergoerselectionareexcluded.(Thelatterapproachwillbepursuedin§6below.)5.3.Whattodowith"symmetricallanguages",i.e.fullyneutralalignment?Aswesawearlier,somelanguageshavebeenreportedasnotmakingagrammaticaldistinctionbetweenRandTinditransitiveconstructions.AnexampleofalanguageinwhichboththeTandtheRarecodedwiththesamecase(Accusative)isthePama-NyunganlanguageMartuthunirathatwesawearlierin(5).Theexamplesin(24)showthatboththeRandtheTmaybepassivized.(24)Martuthunira(Dench1995:229)a.Ngunhupawuluyungku-yangumurla-anganaju-wu-luyaan-tu.that.NOMchildgive-PASS.PFVmeat-ACCI-GEN-EFFwife-EFF'Thatchildwasgivenmeatbymywife.'b.Nhiyumurlayungku-yanguyirnakanyara-anguluwartirra-lu.this.NOMmeatgive-PASS.PFVthis.ACCMann-ACCjene.EFFwoman-EFF'Thismeatwasgiventothismanbythatwoman.'Otherlanguageswith"fullyneutral"constructionsareCavineña(Tacanan;Bolivia;Guillaume2006),thePeruvianPanoanlanguagesShipibo-Konibo(Valenzuela2001)andMatsés(Fleck2001),andHaruai(UpperYuat;PapuaNewGuinea;Comrie1993:317).AnothersuchlanguageisKinyarwandaaccordingtoGary&Keenan(1977).FullneutralityofthissorthasbeenaproblemforRelationalGrammar,whichclaimsthat1,2and3areuniversalrelations.Thus,RandTarepredictedtobedistinguishableinalllanguagesinsomeway,thoughnothingissaidabouthow.AsInotedearlierin§3.1,Gary&Keenan(1977)hadclaimedthatKinyarwandashowsafullysymmetricalditransitiveconstruction,andDryer(1983)rescuedtheRelationalGrammaraccountofKinyarwandabyobservingthattherearesomelesssalientwaysinwhichtheTandtheRdifferafterallinKinyarwanda(withrespecttocausativizationand"locativeadvancement").Unfortunately,thisclaimisimmuneto\n14falsificationinpracticalterms:Thereisnowayonecouldexhaustivelyexamineallpossiblyrelevantconstructionstodeterminewhethertheyprivilegeoneofthetwoarguments,soonecanalwaysclaimthatthereisprobablysomeconstructionwithrespecttowhichRandTdiffer,eventhoughithasn'tbeendiscoveredyet.SymmetryofthissortalsopresentsachallengetoRoleandReferenceGrammar,asitdoesforotherframeworks.Dotheselanguageshavetwoundergoersinditransitiveconstructions?Ordotheyhavenoundergoer(i.e.aretheM-intransitive)?Theformerisexcludedbytheprinciplethatthereareatmosttwomacrorolesperclause(VanValin2005:64),andthelatterishardlyanattractiveoptionbecauseMartuthuniraallowspassivization.VanValin(2007:§4.2)discussessymmetricalpassivizationinKinyarwandaandproposesthatpassivizationdoesnotmakereferencetotheundergoernotioninthislanguage,buttargetsallnon-macroroledirectcorearguments.Hedoesnotsayanythingaboutwhichofthetwonon-ActorargumentsisundergoerinKinyarwanda,andpresumablyinafullyneutrallanguage,therewouldbenoundergoeratall(oratleastnoevidenceforit).Thispositionisprobablyconsistentwiththeoverallframework(afterall,thereareotherthree-argumentverbslackinganundergoer,e.g.talktosomebodyaboutsomething),butintuitivelyitisstrangetoclaimthatditransitiveconstructionsareintransitive.5.4.WhichpropertiesarerelevantforUndergoerselection?Butitseemsthatsomecriteriafordistinguishingargumentsaregivenmoreweightthanothersbysomeauthors.Hudson(1992)notesthatonlypassivizationsupportstheR-as-direct-objectdescriptionoftheEnglishDouble-ObjectConstruction.Asshownby(25b-c),theRcanbetheprivilegedsyntacticargument("subject")ofthepassive,andtheTcannot(orcannotformanyspeakers).(25)a.(monotransitive)PedrowascriticizedbyAisha.b.(passive)R=UMPedrowasgivenmoneybyAisha.c.(passive)T=UM??MoneywasgivenPedrobyAisha.AsHudsonobserves,quiteafewotherbehaviouralpropertiessupportthegroupingofTwithmonotransitiveU.Forexample,boththeUandtheTcanbepivotwithomitted-objectinfinitives:(26)a.(monotransitive)IboughtititoputØionthetable.b.T=UMHegaveherjititoputØionthetable.c.R=UM*HegaveheriitjtocheerØiup.Hudsondoesnotprivilegethepassive,butinsteadcountsthenumberofpropertiesthattreatTandUalike,andfindsthattherearemanymorethanpropertiesthattreatRandUalike.HeconcludesthatitisbettertosaythattheTisthedirectobjectinEnglish.CarryingthisargumentationovertoRRG,couldonesaythatinEnglish,theditransitivethemeistheUndergoer?VanValin(2007:n.2)answersnegatively:Thisisuntenable,forthefollowingreason.TheRRGanalysisofpassiveisthattheundergoerappearsasthePSA('subject')inlanguageslikeEnglish.\n15TheRRGanalysispredictsthatthepassiveacceptabletoallEnglishspeakerswillbeKimwasgiventhebookbyPat,whereastheotheranalysispredictsthattheuniversallyacceptablepassiveformwouldbeThebookwasgivenKimbyPat.Soapparentlypassivizationisaknock-downcriterionforUndergoerselection.Butwhyispassivizationgivensomuchprominence?Passivizabilityisjustoneamongmanypropertiesofobjects,anditwouldseemtobedifficulttoarguethatitshouldbemoreimportantthantheotherargumentproperties7(cf.thequestionablePassivePrivilegeprincipleof§1).6.AsketchofaradicalalternativeRRGaccountToaddresstheissuesraisedabove,theRRGaccountofditransitiveconstructionswouldhavetoberevised.AlthoughIamnotinapositiontoprovideafull-fledgedRRGaccount,Iwillnowsketchtheoutlinesthatsuchanalternativeaccountcouldhave.6.1.Fourmacroroles:A,U,RandTFirstofall,RRGwouldneedtwoadditionalmacroroles,recipient(R)andtheme(T),inadditiontoactor(A)andundergoer(U):(27)foursemanticmacrorolesandtheirdefinitionsA=actor,definedasbeforeU=undergoerofmonotransitivesentencesR=macro-recipient,correspondingtoPrimus's(1999)proto-recipientT=(macro-)themeIntheradicalalternativeaccount,theproblemsdiscussedin§5wouldbeavoided.Thefourmacrorolesaredefinedpurelysemantically,i.e.theradicalalternativeaccountavoidsthemixingofsemanticandsyntacticinformationandobservesdescriptiveautonomy(see§5.2).Thereisno"non-default"or"marked"selectionofmacroroles,sothatthemarkednessproblemsof§5.1areavoided.Nocriterionforgroupingrolesisgiveninherentpriority,i.e.passivizationwouldnotbemoreimportantthanothercriteria(see§5.4).SymmetricalconstructionswithneutralalignmentaresimplyregardedasconstructionswhereU,RandTarecodedinthesameway(see§5.3).Sincethenon-actorargumentsofmonotransitiveandditransitivesentenceshavedifferentmacroroles,thequestionofwhichditransitivenon-actorargumentistheundergoerwouldnotarise.NotealsothattheRRGliteraturealreadycontainsoneargumentforthemacrolrole(macro-)recipient:SuchamacrorolewouldbeusefulfordescribingrecipientpassivesinGerman,asarguedbyDiedrichsen(2004).6.2.Codingrulesandtheparallelwithmonotransitivealignment7PerhapstheimportanceofpassivizationinRRGhastodowiththefactthatRRGaroseinthe1970s,whenmanysyntacticianswereinterestedinpassivization,andallmajorgrammaticaltheories,suchasRelationalGrammar,FG,GPSG,LFG,SpaceGrammar/CognitiveGrammar,feltcompelledtobeginthejustificationfortheirgrammarwithananalysisofthepassive.\n16AnimportantcriterionforanRRGanalysisistheelegancewithwhichrulesforcodingargumentscanbeformulated.Whilethiscriterionofdescriptivesimplicityisproblematic(aswasnotedin§1),itisstillinterestingtoseehowtherulesforcodingelementswouldhavetobereformulatedinthenewsystemwithfourmacroroles.Takingintoaccountthemajoralignmentpatternsofbothmonotransitiveandditransitiveconstructions,wecandistinguishfourmajortypesoflanguages(cf.Dryer2007).Sincetheergativityparameterandthesecundativityparameterareindependentofeachother(Haspelmath2005a),allfourarewellattested(cf.alsoVanValin2005:127).Figure4showsthesefourtypes,withoneexemplifyinglanguageforeachtype.Thelabelsbelowthelanguagenamesarethemacrorolegroupingsthataretreatedalikeinthelanguage(hereonlycase-markingisrelevant;otherconstructionsmaybehavedifferently).accusativealignmentergativealignmenttSnomSabsnemnAUaccAUgliaeTRTdirRivdircteirGermanLezgiand(nominativeand(absolutive-directive)inaccusative-directive)tnSnomSabsemgnAUaccAUliaprievTRpriTRtiadnYorubaGreenlandiccue(nominativeand(absolutive-primative)saccusative-primative)Figure4:Fourtypesoflanguages,accusativeandergativealignmentTodescribetherulesforcases,RRGappealstoasimpleMacroroleHierarchy"actor>undergoer".Intheradicalalternativesystem,withfourmacroroles,themacrorolehierarchywouldbeasin(28):(28)MacroroleHierarchyA>R,U>TThatis,actorishighestonthehierarchy,andditransitivethemeislowest.Ditransitiverecipientandmonotransitiveundergoerarebothintermediate(andnotrankedwithrespecttoeachother).Thecase-markingrulesoftraditionalRRGaregivenin(29)(VanValin&LaPolla1997:§7.3).(NotethatMRstandsformacrorole.)(29)a.(accusativealignment)\n17Nominativecaseisassignedtothehighest-rankingMRargument,thenaccusativetotheother.b.(ergativealignment)Absolutivecaseisassignedtothelowest-rankingMRargument,thenergativetotheother.Thisisaverysimpleandstraightforwardsystem,andtheadditionalrulefordativecase(itisthedefaultcase)couldnotbesimpler.Bycontrast,thecase-markingrulesoftheradicalalternativemacroroleapproachwouldbesomewhatmorecomplex,asshownin(30).(30)a.(German-stylealignmentcombination:accusative-indirective)Accusative-directivecaseisassignedtothelowest-rankingMRargument,thennominativeisassignedtothehighest.b.(Yoruba-stylealignmentcombination:accusative-secundative)Accusative-primativecaseisassignedtothesecondhighest-rankingMRargument,thennominativetothehighest.c.(Lezgian-stylealignmentcombination:ergative-indirective)Absolutive-directivecaseisassignedtothelowest-rankingMRargument,thenergativetothehighest.d.(Greenlandic-stylealignmentcombination:ergative-secundative)Absolutive-primativecaseisassignedtothesecondlowest-rankingMRargument,thenergativetothehighest.Thisisnotdramaticallymorecomplicated,anditaccountsforalotmorefacts8thanthetraditionalRRGcase-markingrules.6.3.Objectionsagainst"athirdmacrorole"ThepossibilityofpositingathirdmacrorolehascomeupearlierintheRRGliterature.VanValin(2004:79-81,2005:64-66)arguesthatathirdmacroroleshouldnotbeposited,foranumberofreasons(seealsoBellostavonColbe2004:194-198).Idonotfindthesereasonscompelling,andsomeofthemmaybeweakenedbythecurrentproposal,whichactuallypositsfourratherthanthreemacroroles.VanValin'sobjectionstoathirdmacroroleare:(i)Athirdmacrorolewouldnotbeuniversal.Thismaybeso,butitisnotclearwhyuniversalityshouldbeacriterion.Ifthetheoryisnon-aprioristic(see§1),itshouldevenallowforthepossibilityofdifferentmacrorolesindifferentlanguages.Moreover,actorandundergoerarealsosomewhatvariableacrosslanguages.Andinthecurrentproposal,thisargumenthaslittleforcebecausemacrorolesarepurelysemanticandthereforeareuniversalbydefinition(assumingthatmeaningisuniversal,atleastatsomelevel).(ii)RandTarenottreatedconsistentlyacrosslanguages,incontrasttoAandU,whicharealways"direct(core)arguments".TheproblemwiththisargumentisthatitisnotclearthatAandUarereallysignificantlydifferent8However,asRobertVanValinpointsout(p.c.),theydonotextendautomaticallytointransitiveverb,sointhisregardtheyareadmittedlylessgeneral.\n18fromRandT.Theconcept"direct(core)argument"isnotdefinedverywell.InVanValin&LaPolla(1997:29)itissaidtobeanargumentthatisnotadpositionallymarked,butpresumablyinstrumentalandlocativeargumentswouldnotcountas"direct(core)arguments".VanValin(2005:7)ismoreprecise:Nominative/absolutive,accusative/ergativeanddativecountas"direct",whereasallothercasesareregardedas"oblique".Noprincipledreasonisgivenforthis,andnodefinitionof"dative"isgiven.(iii)"AthirdmacrorolewouldbemarkedlylessimportantforthesyntaxthanAandU...Italsoplaysnoroleinthemajortypologyofsyntacticsystems:ergativevs.accusativevs.split-intransitive."(VanValin2004:81)Thisisamatterofperspective.ItistruethattherolesRandTarelessimportantthanAandUinthestraightforwardsenseofoccurringlessfrequentlyintexts.Butotherwisetheparallelsbetweenmonotransitiveand9ditransitivealignmentarestriking,andintraditionalRRGtheaccusative/ergativecontrastistreatedinawaythatisverydifferentfromtheindirective/secundativecontrast.ProbablythemostseriousobjectiontotheradicalalternativethatIhaveproposedisthatitwouldmakeanalysesofotherphenomenathatIhavenotdiscussedheremorecomplicated(e.g.ofcausativeconstructions).Thismaywellbe,butitwouldnotnecessarilybeanargumentagainsttheproposal.Recallfrom§1thatdescriptivesimplicityisacriterionthatIdonotregardasparticularlyimportant,becausewehavenostrongreasontoassumethatlanguagesweredesignedtobesimple.Ontheotherhand,theothercriteria(andnon-criteria)of§1,takentogether,clearlyfavorthealternativeoverthetraditionalaccount.Thus,whetherthealternativeisadoptedwillultimatelydependonhowonerelatestotheprinciplesof§1.IwouldhopethatfutureworkonRRGwillatleastclarifywhereRRGstandsonthesematters.Theremainderofthispaperwilldiscusstwofurtherissues:TherevisedRRGapproachintroducedbyGuerrero&VanValin(2004)andVanValin(2007),andthepossibilityofasecondgrammaticalrelationforditransitiveconstructions.7.TherevisedRRGaccount(VanValin2007)Guerrero&VanValin(2004)andVanValin(2007)recognizethattheRRGapproachtoditransitivesassetoutbyVanValin&LaPolla(1997)doesnotsufficientlytakeintoaccountlanguageswithsecundativealignmentandthusviolatestheprincipleoftypologicaladequacy(see§1).Theydiscussonesuchlanguage,Yaqui(aUto-AztecanlanguageofMexico),insomedetail,showingthatthesimplestdescriptionforYaquiisoneintermsoftherulethattheundergoeristhesecondhighestrankingargumentintheLogicalStructure.ThisisinconflictwiththeuniversalmarkednessprincipleofFigure1,whichtheyabandoninfavourofa"parameterized"systemforundergoerselection.ThisissummarizedinFigure4.9Evena"split-transitive"type(cf.Siewierska2003)hasbeenidentified,analogoustothesplit-intransitivetypeinmonotransitivealignment.\n19Figure4YaquifollowsPrincipleB(likeothersecundativelanguages),GermanfollowsPrincipleA(likeotherindirectivelanguages).(SeealsoContiJiménez2004,whoarguesthatthisaccountsforthecross-linguisticfacts.)Thisapproachdoesmorejusticetothesymmetryoftheindirectiveandthesecundativealignmentpatterns,andthusaddressestheproblemsdiscussedin§5.1.Buttherearestillanumberofproblems:(i)Undergoerselectionisnowparameterizedfortheindirective-secundativecontrast,inmuchthesamewayassometheoriesparameterize"subjectselection"fortheaccusative-ergativecontrast.ThisisrejectedbyRRG,butnowthattheparameterizationoptionhasbeenallowedintoRRG,onecouldaskwhythecross-linguisticvariationinmonotransitiveconstructionsisnottreatedinthesameway.(ii)RRG'smacrorolesactorandundergoerweredesignedtocapturewhataccusativeandergativesystemshaveincommon—therearenoPrinciplesA(foraccusativelanguages)andB(forergativelanguages).Giventheparallelsbetweenmonotransitiveandditransitivealignment,itseemsmoreinthespiritofRRGtopositmacrorolesalsoforditransitivealignment.ThenewRRGaccountthusdoesnotexpresstheparallelsbetweenmonotransitiveandditransitivealignmentsinthesamestraightforwardwayastheradicalalternative.8.DoweneedagrammaticalrelationforRandT?AsexplainedindetailinVanValin&LaPolla(1997:ch.6),VanValin(2005:89-107),grammaticalrelationsarepositedbyRRGwhenthereisarestrictedneutralizationofsemanticrolesformorphosyntacticpurposes.Thisisthecaseinanumberofconstructionsforwhichaprivilegedsyntacticargument(PSA)hasbeenposited.Roughly,thePSAcorrespondstowhathasbeencalled"subject"inotherapproaches.Sincetheseotherapproachesoftenalsopositagrammaticalrelation"object",onemightwonderwhetherananalogousconceptofan"SSA"(SecondarySyntacticArgument)mightbeneededinRRGaswell.VanValin(1993:65-72)arguesthatnosuchconceptisneeded,andthatalltheworkdonebyan"object"relationinothertheoriesisdonebytheundergoermacroroleandotherRRGnotions.However,notallgroupingsofnon-Actorargumentscanbecapturedbythemacroroles.Inparticular,oneandthesameconstructionmaysometimesshowboththeU+Tgrouping(i.e.directive,or"directobject")andtheU+Rgrouping(i.e.primative,or"primaryobject").AnexamplecomesfromHyow,aTibeto-BurmanlanguageofBangladesh:\n20(31)Hyow(Peterson2003:174,179)a.yɔntɯʔauy=lakeyʔɔ-ŋoʔwey-sɔyesterdaydog=ERGI[ABS]1SG.U/R-bite-CONCLAU'Yesterdayadogbitme.'b.cu=lakey=acɔʔe-pekhe=ERGI=LOCbook[ABS]1SG.U/R-giveART'Hegavemeabook.'Herecase-markingshowsanindirectivepattern(U+Tiszero-coded,andRisinthelocative),whileindexing("agreement")issecundative(U+Rareindexedovertly,andTisnotindexed).IntraditionalRRG,thiscannotbedescribedintermsofmacroroles,unlessoneadmitsthataclausegetstwoundergoers,one"A-undergoer"forcase-marking(assignedbyPrincipleA),andone"B-undergoer"forindexing(assignedbyPrincipleB).AsimilarexamplecomesfromEnglish,wherePassivizationandOmitted-ObjectInfinitivesbehavedifferently,asobservedbyHudson1992(cf.alsotheearlierdiscussionin§5.4):Passivization:secundative(U+R(passivizable)vs.T(notpassivizable)(32)a.(monotransitive)PedrowascriticizedbyAisha.b.(passive)R=UMPedrowasgivenmoneybyAisha.c.(passive)T=UM??MoneywasgivenPedrobyAisha.PivotwithOmitted-ObjectInfinitives:indirective(U+Tvs.R)(33)a.(monotransitive)IboughtititoputØionthetable.b.T=UMHegaveherjititoputØionthetable.c.R=UM*HegaveheriitjtocheerØiup.Thus,intermsofgrammaticalrelations,itappearsthatweneedtosaythatinEnglish,thecontrollerofPassivizationisU+R,whilethepivotofOmitted-ObjectInfinitivesisU+T.Ofcourse,inRRGonecouldalwaysresorttoanalysesthatdonotmakereferencetomacroroles(e.g.onecouldsaythatthepivotofOmitted-ObjectInfinitivesisthelowestrankingcoreargument,asRobertVanValinpointsout,p.c.),butitseemsthatthereisnotalwaysaprincipledwayofdecidingwhenreferenceshouldbemadetomacrorolesandwhenitmacrorolesshouldbeleftaside.Inanyevent,whiletheremaybenoconstructionrequiringtwodifferentprivilegedsyntacticargumentsatthesametime(whichwouldleadtotheneedofanSSAinadditiontoaPSA),itdoesseemthatalsowithrespecttogrammaticalrelations,ditransitiveconstructionsbehavemuchlikemonotransitiveconstructions.9.Conclusion:Shiftingthesyntax-semanticsboundaryinfavourofsyntaxInconclusion,IwouldliketoproposethatRRGshouldbecomelessaprioristicandshouldseparatesyntaxandsemanticsmorestrictly:the\n21macrorolesshouldbestrictlysemantic,andcorrespondingly,somewhatmoreworkshouldbedonebysyntax.Assuggestedbythediscussionin§8,weseemtoneedgrammaticalrelationsotherthanPSA.ItmustbeadmittedthatiftheapproachfavoredhereisadoptedbyRRGpractitioners,theresultingdescriptionswilloftenbelesselegantandmorecomplexthanthoseoftraditionalRRG.However,Idonotbelievethatitisavirtueofgrammaticaltheoriesiftheyallowlinguiststoformulatesimpledescriptionsoflanguages(seethediscussionofdescriptivesimplicityin§1),atleastnotifthiscomesattheexpenseofacomplicatedarchitecturewithmacrorolesthathaveamixedsemantic-syntacticmotivation.Itneedstobeemphasizedthatseparatingsyntaxandsemanticsinthewayadvocatedheredoesnotmeanacceptingformalistautonomy.Descriptiveautonomyofthesemanticsandsyntaxisfullycompatiblewithfunctionalism(asarguedbyNewmeyer1998,andshownbythefactthatthisauthorhasgoneontopursueafullyfunctionalistagendainNewmeyer2005).Adoptingthealternativeapproachproposedherewouldmeanarapprochementwiththenon-aprioristicapproachesofDryer(1997)andCroft(2001),bothofwhichemphasizethatlanguagesarefarmorediversethanisgenerallyrecognized(thusrequiringextensivelanguage-particularstipulation),butthatwhatevergeneralityexistsisamenabletoexplanationinsemantic-pragmaticterms.ReferencesBellostavonColbe,Valeriano.2004."DasindirekteObjektalssyntaktischesArgumentohneMakrorolle."In:Kailuweit,Rolf&Hummel,Martin(eds.)SemantischeRollen.Tübingen:Narr,184-205.Blake,BarryJ.1990.Relationalgrammar.London:Routledge.Blansitt,EdwardL.Jr.1984."Dechticaetiativeanddative.In:In:Plank,Frans(ed.)Objects.London:AcademicPress,127-150.Comrie,Bernard.1993.“SomeremarksoncausativesandtransitivityinHaruai”.InCausativesandTransitivity,Comrie,BernardandMariaPolinsky(eds.),315ff.ContiJiménez,Carmen.2004."Beneficiaryandrecipientindoubleobjectconstructions."ProceedingsoftheRRG'04Conference,110-119.(availablefromRRGwebsite,http://wings.buffalo.edu/linguistics/rrg/)Croft,William.1990.Typologyanduniversals.Cambridge:CambridgeUniversityPress.Croft,William.1995.“Autonomyandfunctionalistlinguistics.”Language71:490-532.Croft,William.2001.RadicalConstructionGrammar.Oxford:OxfordUniversityPress.Dench,AlanCharles.1995.Martuthunira:ALanguageofthePilbaraRegionofWesternAustralia.PacificLinguistics,SeriesC-125.Canberra:Dept.ofLinguistics,ResearchSchoolofPacificandAsianStudies,AustralianNationalUniversity.Diedrichsen,Elke.2004."TheGerman'bekommen-passive'andRRG."ProceedingsoftheRRG'04Conference,49-71.(availablefromRRGwebsite,http://wings.buffalo.edu/linguistics/rrg/)Dik,SimonC.1989.ThetheoryofFunctionalGrammar.Dordrecht:Foris.Dik,SimonC.1997.ThetheoryofFunctionalGrammar,vol.1-2.Berlin:MoutondeGruyter.Dryer,MatthewS.1983."IndirectobjectsinKinyarwandarevisited."In:Perlmutter,DavidM.(ed.)StudiesinRelationalGrammar1.Chicago:TheUniversityofChicagoPress,129-140.Dryer,MatthewS.1986."Primaryobjects,secondaryobjects,andantidative."Language62:808-45.Dryer,MatthewS.1997."Aregrammaticalrelationsuniversal?"In:Bybee,Joan&Haiman,John&Thompson,SandraA.(eds.)Essaysonlanguagefunctionandlanguagetype.Amsterdam:Benjamins,115-143.Dryer,MatthewS.2007."Clausetypes."In:Shopen,Timothy(ed.)Languagetypologyandsyntacticdescription,2nded.Cambridge:CambridgeUniversityPress.\n22Fleck,DavidW.2001.CausationinMatses(Panoan,AmazonianPeru).InThegrammarofcausationandinterpersonalmanipulation.MasayoshiShibatani(Ed.).Amsterdam:Benjamins,373-415.Frajzyngier,Zygmunt.2006."Towardsanon-aprioristicsyntax:grammaticalandsemanticrelationsunderconstraintsonform."Plenarytalkatthe2ndSyntaxoftheWorld'sLanguagesconference,LancasterUniversity,September2006.Gary,JudithOlmstedandEdwardKeenan.1977.'OncollapsinggrammaticalrelationsinUniversalGrammar.'In:SyntaxandSemantics8,83-120.Goldberg,AdeleE.1992."Theinherentsemanticsofargumentstructure:thecaseoftheEnglishditransitiveconstruction."CognitiveLinguistics3:37-74.Goldberg,AdeleE.1995.Constructions:AConstructionGrammarapproachtoargumentstructure.Chicago,TheUniversityofChicagoPress.Guerrero,Lilián&VanValinJr.,RobertD.2004."Yaquiandtheanalysisofprimaryobjectlanguages."InternationalJournalofAmericanLinguistics70.3:290-319.Guillaume,Antoine.2006."Ditransitivitéencavineña:constructionsàobjetdouble."UniversitédeLyon(toappearinAmerindia)Harley,Heidi.2002.“Possessionandthedouble-objectconstruction.”In:Pica,Pierre&Rooryck,Johan(eds.)LinguisticVariationYearbook2.Amsterdam:Benjamins,31-70.Haspelmath,Martin.2000.“Whycan’twetalktoeachother?Areviewarticleof[Newmeyer,Frederick.1998.Languageformandlanguagefunction.Cambridge:MITPress.]Lingua110.4:235-55.Haspelmath,Martin.2005a."Argumentmarkinginditransitivealignmenttypes."LinguisticDiscovery3.1:1-21(freeonlinejournal,http://linguistic-discovery.dartmouth.edu/)Haspelmath,Martin.2005b."DitransitiveConstructions:TheVerb‘Give’."In:MartinHaspelmath&MatthewS.Dryer&DavidGil&BernardComrie(eds.)TheWorldAtlasofLanguageStructures.Oxford:OxfordUniversityPress,426-429.Haspelmath,Martin.2006."Againstmarkedness(andwhattoreplaceitwith)."JournalofLinguistics42.1:25-70.Haspelmath,Martin.2007a."Ditransitivealignmentsplitsandinversealignment".ToappearinFunctionsofLanguage(specialissueonditransitives,guesteditedbyAnnaSiewierska)Haspelmath,Martin.2007b."Pre-establishedcategoriesdon'texist:consequencesforlanguagedescriptionandlanguagetypology."ToappearinLinguisticTypologyHudson,Richard.1992."Socalleddoubleobjectsandgrammaticalrelations."Language68:251-276.Kailuweit,Rolf.2005.LokativalternanzbeitransitivenVerben:Englisch,Französisch,SpanischundDeutschimVergleich.In:ChristianSchmitt&BarbaraWotjak(eds.)Beiträgezumromanisch-deutschenundinnerromanischenSprachvergleichen.Bonn:RomanistischerVerlag.Manning,Christopher.1996.Ergativity:argumentstructureandgrammaticalrelations.Stanford:CSLI.Newmeyer,FrederickJ.1998.Languageformandlanguagefunction.Cambridge,MA:MITPress.Newmeyer,FrederickJ.2005.Possibleandprobablelanguages.Oxford:OxfordUniversityPress.Perlmutter,DavidM.1980."RelationalGrammar."InMoravcsik,EdithA.&JessicaR.Wirth.1980.Currentapproachestosyntax(Syntaxandsemantics13).NewYork:AcademicPress,195-229.Peterson,DavidA.2003."AgreementandgrammaticalrelationsinHyow."In:Bradley,David&LaPolla,Randy&Michailovsky,Boyd&Thurgood,Graham(eds.)Languagevariation:PapersonvariationandchangeintheSinosphereandintheIndosphereinhonourofJamesA.Matisoff.(PacificLinguistics,555.)Canberra:AustralianNationalUniversity,173-83.Peterson,DavidA.2007.Applicativeconstructions.Oxford:OxfordUniversityPress.Primus,Beatrice.1999.Casesandthematicroles:ergative,accusativeandactive.Tübingen:Niemeyer.RappaportHovav,Malka&Levin,Beth.2006."TheEnglishdativealternation:thecaseforverbsensitivity."Ms.,StanfordUniversity.Rohdenburg,Günther.1996."CognitivecomplexityandincreasedgrammaticalexplicitnessinEnglish."CognitiveLinguistics7.149-182.Rowlands,E.C.1969.Yoruba.(TeachYourselfBooks)Sevenoaks,Kent:Hodder&Stoughton.Siewierska,Anna.1998.Languageswithandwithoutobjects.LanguagesinContrast1.2:173-190.\n23Siewierska,Anna.2003."Personagreementandthedeterminationofalignment."TransactionsofthePhilologicalSociety101.2:339-370.Valenzuela,PilarM.2001.CausativizationandtransitivityinShipibo-Konibo.InThegrammarofcausationandinterpersonalmanipulation.MasayoshiShibatani(Ed.).TSL49.Amsterdam/Philadelphia:JohnBenjaminsPublishingCompany.pp.417-483.VanValin,RobertD.,Jr.1993."AsynopsisofRoleandReferenceGrammar."InVanValin(ed.)AdvancesinRoleandReferenceGrammar.Amsterdam:Benjamins,1-164.VanValin,RobertD.,Jr.2004."SemanticmacrorolesinRoleandReferenceGrammar."In:Kailuweit,Rolf&Hummel,Martin(eds.)SemantischeRollen.Tübingen:Narr,62-82.VanValin,RobertD.,Jr.2005.Exploringthesyntax-semanticinterface.Cambridge:CambridgeUniversityPress.VanValin,RobertD.,Jr.2006."Patternsofrestrictedneutralizationingrammaticalrelationssystems."Plenarytalkatthe2ndSyntaxoftheWorld'sLanguagesconference,LancasterUniversity,September2006.VanValin,RobertD.,Jr.2007."TheRoleandReferenceGrammaranalysisofthree-placepredicates."SuvremenalingvistikaVanValin,RobertD.,Jr.&LaPolla,RandyJ.1997.Syntax:structure,meaningandfunction.Cambridge:CambridgeUniversityPress.Wunderlich,Dieter.1997."Causeandthestructureofverbs."LinguisticInquiry28:27-68.Wunderlich,Dieter.2006.Towardsastructuraltypologyofverbclasses.In:Wunderlich,Dieter(ed.)2006.Advancesinthetheoryofthelexicon.Berlin:MoutondeGruyter,57-166.