A Formalist’s Reading of Some Functionalist Work in Syntax Stephen R. Anderson Dept. of Linguistics, Yale University Abstract In terms of social groups, “formalist” syntacticians are roughly those who follow some variant of the program of Generative Grammar (Princi- ples & Parameters, LFG, HPSG, Minimalism, etc.). Their analyses tend to be based on assumptions of Modularity (including the “Autonomy of Syn- tax” as a special case) and categoriality, among other principles. In more general terms, however, it can be argued that “formalism” simply consists in a commitment to fully explicit formulations cashing out one’s intuitions about the structure of language in terms that require as little as possible in the way of unanalyzed contributions by an understanding reader: surely a ‘motherhood’ issue that could not plausibly differentiate theoretical views. “Functionalists,” typically, are those who argue for a higher degree of in- volvement of other domains (semantics, pragmatics, discourse, extra-linguistic exigencies deriving from the context of communication, etc.) in syntactic phenomena, and for hierarchies, gradients, and other non-categorial anal- yses. I argue, however, that the practice of many functionalist syntacti- cians generally trades heavily on a relatively low degree of explicitness and on pre-systematic, intuitive understandings of the categories of an analysis. When functionalist arguments against modularity, or in favor of hierarchi- cal scales as opposed to discrete categories are examined closely, they often break down on just the basis that they involve assumptions about the unity of domains of fact that are better seen as the product of distinct interacting systems. The activity of examining functionalist arguments in this way is of- ten quite instructive, but not always in the direction their formulators might have intended.
Transcript
A Formalist’sReadingof SomeFunctionalistWork in Syntax
StephenR. Anderson�
Dept.of Linguistics,YaleUniversity
Abstract
In termsof social groups,“formalist” syntacticiansare roughly thosewho follow somevariantof the programof Generative Grammar(Princi-ples& Parameters,LFG, HPSG,Minimalism,etc.). Their analysestendtobebasedon assumptionsof Modularity (including the “Autonomyof Syn-tax” asa specialcase)andcategoriality, amongotherprinciples. In moregeneralterms,however, it canbe arguedthat “formalism” simply consistsin a commitmentto fully explicit formulationscashingout one’s intuitionsaboutthestructureof languagein termsthat requireaslittle aspossibleinthe way of unanalyzedcontributionsby an understandingreader:surelya‘motherhood’issuethatcouldnotplausiblydifferentiatetheoreticalviews.
“Functionalists,” typically, arethosewhoarguefor ahigherdegreeof in-volvementof otherdomains(semantics,pragmatics,discourse,extra-linguisticexigenciesderiving from the context of communication,etc.) in syntacticphenomena,and for hierarchies,gradients,andothernon-categorial anal-yses. I argue, however, that the practiceof many functionalistsyntacti-ciansgenerallytradesheavily ona relatively low degreeof explicitnessandon pre-systematic,intuitive understandingsof thecategoriesof ananalysis.Whenfunctionalistargumentsagainstmodularity, or in favor of hierarchi-calscalesasopposedto discretecategoriesareexaminedclosely, they oftenbreakdown on just thebasisthat they involve assumptionsabouttheunityof domainsof fact thatarebetterseenastheproductof distinct interactingsystems.Theactivity of examiningfunctionalistargumentsin thiswayis of-tenquiteinstructive, but not alwaysin thedirectiontheir formulatorsmighthave intended.
My (assigned)topic hereis the questionof what formalistscan learn fromfunctionalistsin syntax. “What canwe learn.. . ” might be interpretedas: whatanalyseshave we seenthatgave usideas?Naturally, whenonereadsthework ofany otherlinguist who caresaboutthelinguistic materialunderdiscussion,therearelikely to bedescriptivepointsthatwill beof interest,but noparticularbroaderpurposewould beservedby anenumerationof casesin which I personallyhavefoundexamplesfor my own argumentsin functionalistsources.
Anotherway of looking at the questionwould be to ask: what pointshavefunctionalistsmadethat would causeme to abandona formalist program?I’mafraidthathereI have to reportthatnothingI have readin thefunctionalistliter-aturehasconvincedme of any suchthing. Certainlytherearephenomenatherethatstill wantexplanations,but nothingsuggeststo methat theseby themselveswarrantbasicrevisions in methodology, asopposedto the variousrevisionsofparticularanalysesthatoughtto beadoptedin specificcases.
While thewayI will addressthequestionin thispaperwill inevitablyhavecer-tain autobiographicalaspects,I will attemptto formulatethe issuein somewhatmoregeneralterms.WhenI readthework of a carefulfunctionalist,how wouldI addressthepointsmade?Thatis, wherefunctionalistsciteempiricaldata,espe-cially datathatareintendedto challengethevalidity of otherpointsof view, I aminterestedin theextentto whichthosedataactuallybearon issuesin thetheoryofformalgrammar, andwherethey do,how. In caseswherethebearingon issuesofprincipleis lessthanwhatis claimed,onealsowantsto know whetherthis resultssimply from poorly constructedarguments,or whetherit reflectssomebroaderprinciplethatcharacterizesanentireapproachto language.In theprocessof con-sideringthosequestions,I think I canlearnsomethingaboutwhat formalismisabout,aboutwaysin which thefunctionalistprogramshows lessthanit purportsto, andaboutsomebroadmethodologicaldifferencesin linguists’ approachestothesubjectmatterof ourdiscipline.�I am grateful to the participantsin the Mil waukeeconferencefor usefulcommentson the
initial oral presentationof this paper, andto Edith Moravcsik andMichael Darnell for detailedremarkson anearlierwritten version.I would alsolike to acknowledgethehelpof my colleagueLarry Horn for generalcomments,andfor invoking his network of friendsandcorrespondentsinclarifying thecorrectattributionof theepistemologicalprinciplein (1).
2
1 What is a Formalist?
In termsof the sociologyof the field, I imaginesomeof my “formalist” friendswould considermea somewhatmarginal memberof their fraternity, andthefirstpointI wantto makemaysolidify thatimpression.In particular, I haveratherseri-ousdoubtsabouttheultimatelyproductivenatureof someimportantassumptionsin recentformal syntax,andabouta style of argumentthat givesrise to them.Some“formalist” work seemsto meto bedrivenby just exactly thewrongsenseof “formalism”: that is, formalismfor its own sake, anapproachto thefield thatallowslinguisticresearchto bedrivenby theæstheticsof anotation.It is onethingto let theconsequencesof one’s formalizationsuggesthypothesesfor exploration— it is quiteanotherto actasthoughthosehypotheseswerethemselvesempiricalresults.
To cite anexamplewheremy opinionwill offenda goodnumberof syntacti-cianswhosework I otherwiseadmire,considerthe foundationfor thewholesalereplacementsomeyearsago of the traditional notationfor clausalstructureintermsof thecategoriesSand
�by anotationin termsof abstractfunctionalheads.
Originally, this consistedin the replacementof S by (someprojectionof) INFL;subsequently, extendingthe sameline of argumentat the instigationof Pollock(1989),INFL wasitself supersededby ahostof AGR’s, T’s,ASP’s,etc.by wayofthe“explodedInfl” hypothesis.To call this a “hypothesis”hasrapidly becomeamisnomer:it is, rather, thebasicworking assumptionwith which beginningstu-dentsareprovidedfor thediscussionof syntacticstructure.Theresultof this lineof thoughtis a climatein which theburningquestionsfor formalistsyntacticianshave cometo be oneslike “Is AGRsP above TP or below it?” ratherthan “Issubjectagreementasyntacticallyautonomousconstituentof representationseventhoughit formspartof asinglewordwith themainVerb?”
Whenwe askwhatthebasiswasfor thewholecottageindustryof functionalheadsin clausalstructure,we can,I think, traceit backto Chomsky’s discussiononp. 3 of Barriers(1986),which I cite in its entirety:
Doesthis system[ � -theoryasdevelopedfor the primary lexicalcategories— sra] extendto the nonlexical categoriesaswell? Evi-dently, theoptimalhypothesisis thatit does.Let usassumethis to becorrect.Thentheclausalcategoriesconventionallylabeled� and ���mightbe ��� � and � � , respectively, where��� INFL and � comple-mentizer.
leapinto the abyss,but it quickly led to a proliferationof structuralarticulationanda reformulationof the natureof syntacticstructurefor which the empiricalfoundationalwaysseemedratherweak to a disinterestedbystander. In fact, inChomsky’s own writing onefindsnumerousargumentsin this samestyle,placeswherea decisionis madeto pursuea certainsetof assumptionsbecausethat iswhat is suggestedby thepropertiesof the formalismof themoment,ratherthanbecauseof a strongintuition that the empiricalgroundingof theseparticularas-sumptionsis strong. Sometimesfollowing this path leadsto quite genuinein-sights,but at othertimesconsiderabledistractionanddetourhasresulted.In thecaseof the ‘explodedINFL hypothesis,’ it is only quite recently, with the Mini-malistprogram(Chomsky 1995andelsewhere),work of EdwinWilliams (1994),my own work on inflectionalmorphosyntax(Anderson1992),andothers,thattheAGR’s have begun to atrophyandthe depthof complexity of positedfunctionalstructurehasbegunto bereducedagain,thoughin at leastsomecases,I fearthisis simply becausethe æstheticshave changed,not becausethe issueshave beenrethoughtongenuinelyempiricalgrounds.
As is presumablyevidentfrom thetoneof theseremarks,I donotbelievethatformalismof this extremely‘pure’ sort, insulatedfrom groundingin mundanelyempiricalconsiderations,is to beencouraged.Of course,if a formalismis reallyservingits purpose,it shouldsuggestlines of inquiry to pursue,but suchsug-gestionsshouldnot beconfusedwith marchingorders.Whatever thesimilaritiesbetweenlinguisticsandmathematics,oursis not a sciencein which creative ele-gancealoneconstitutesa significantresult. Of course,letting thepropertiesof aformalismsuggestitemsfor a researchagendahasoftenprovento bea produc-tive strategy, andI certainlywould not claim that the consequencesof pursuingthestudyof functionalcategoriesin theway thatgrew out of Chomsky’s remarkabovehasbeena wasteof time. But let uskeepour prioritiesstraight:theaim oflinguisticsis insightinto thenatureof language,notelegancefor its own sake.
Ratherthanconfusingformal elegancein itself with empiricalresults,anal-ternativeconceptionof theroleof formalismin Linguisticsis to seeit assimplyacommitmentto explicitness,awayof fully explicatingthestructurewebelievewefind in language.As a formulationof whatI amgettingathere,I wasquitetakenwith someremarksI heardrecentlyfrom JacobLurie, the high schoolstudentwho won first prizein the1996WestinghouseScienceFair. Theyoungmanwasbeinginterviewedby areporterfrom NPR;sincetheprizewasfor hiswork onthecomputationalpropertiesof “surrealnumbers,” theinterviewer tried to gethim totalk aboutjust whatthosewerefor a while, until it becameclearthattheanswerswerenot turninginto greatradio,andhethenshiftedthetopic to themoregeneral
4
questionof justwhatkind of activity mathematicsis. I donothavea transcriptofexactlywhatMr. Lurie said,but it wasroughlythefollowing:
“What do you do whenyou do mathematics?You think aboutsomekind of object,andyoudevelopyourintuitionsabouttheobject.Thenyou try to expressthoseintuitionsin termsof a formal system.Thenyouexplorethepropertiesof thatsystem,to seeif they reallydocorrespondto the intuitionsyou hadaboutthekind of objectyou’retrying to understand.”
Thepoint hereis just theonethatoughtto bemadeaboutformalismin Lin-guistics:thegoalof a formalizationis not elegancein itself, but ratherassistancein cashingout one’s intuitions explicitly. What may have a chanceto countasinsightsandresultsarereally theseintuitions:ourunderstandingof whatit is thatis systematicandcoherentin theworkingsof language.As scientists,we wanttoexplicatetheseintuitionsasfully aspossible,with nothingleft to theimaginationor creativeunderstandingof thereader. In thissense,acommitmentto formaliza-tion would seemto bea ‘motherhood’issue.Everyonehasto wantto beexplicit— that’s justpartof whatmakeswhatwe’re doing“science.”
Or soyou’d think. As Croft (1991:275)observes,“somefunctionalists(e.g.,GeorgeLakoff) [. . . ] arguethat their theoriesshouldnot andevencannotbesoformalized.” LikeCroft, I will passoverthispositionwithoutmuchcomment.ButI do think thata greatdealof work within the functionalistapproach,or at leastsomeimportantexemplarsof that approach,tradesin essentialwayson inade-quatedegreesof formalizationin this genericsenseof explicitness.In particular,thenotionthatdescriptivecategories,andespeciallythoseof traditionalgrammat-ical description,aresomehow given,so asto definethe objectof inquiry in thestudyof language,ratherthanthemselvesrequiringformalexplicationandrecon-struction,seemsto meto underliesomepoorly foundedfunctionalistcriticismsofthe activities of targetedformalists. I want to stressthat this particularproblemdoesnotseemto meto beinherentin thefoundationsof thefunctionalistagenda.It is moreof a lifestyle thana “virtual conceptualnecessity,” but it doesappeartobeendemicin certaincircles.
2 How haveFunctionalists influencedFormalists?
Functionalismitself, asanapproachto grammar, meansratherdifferentthingstodifferentpeople.Croft (1995a)givesasurvey of diverseschoolsof functionalism
5
thatseemsgenerallyfair to me.In his terms(Croft 1995a,p. 490),thebasicpointis that “[f]unctional analysesof grammar[. . . ] centeron linguistic explanationbasedon language’s function in a larger context.” That is, they argue that thenatureof what languageis follows from what speakersdo with it. Differencesamongfunctionaliststhen tendto centeron the extent to which they insist thatis not even possibleto formulatethe structuralpropertiesof languagewithoutessentialreferenceto mattersthatareoutsideof thesystemof languageitself.
To designatetwo moreor lessopposingcommunitiesof syntacticians,then,ifthe“formalists”arethosewhoarecommittedtonotionsinvolvingtheautonomyofthesyntacticsystem,or theappropriatenessof amodularanalysiswhichabstractsvariouspartsof thesystemfrom othersandanalyzesthemin self-containedwaysasindividual contributorsto a morecomplex overall reality, the “functionalists”arethosewhowoulddeny thebasicadequacy of sucha pointof view.
At oneendof Croft’sspectrumof functionalistsarepeoplelikeSusumuKuno,Ellen Prince,andmy colleagueLarry Horn,who do thingsrathersimilar to whatformalistsdo,with thedifferencethattheiranalysesoftenmakeexplicit referenceto meaning,pragmatics,anddiscoursefunction.Work of thissort,it seemsto me,challengesnot somuchthe basicnotionof modularityin grammarasthe actualboundariesof the internally coherentmodules. Perhapsthat is why it doesnotfigureprominentlyin thepolemicsof eitherside.
On theotherhand,perhapspreciselybecauseof thesimilarity in assumptionsand stylesof argument,it is fairly easyfor non-functionaliststo learn from acloseexaminationof this kind of work. For example,a seriesof worksby Kuno(1972,1987,Kuno & Kaburaki 1977)datingbacksomeyearsmadeit clear(tothosewho listened)that a unitary, purely structuralcondition(suchasPrincipleA of the binding theory)could not be developedto accountfor all instancesofreflexivepronounsevenin English,let alonein all naturallanguages.Kunoarguedfrom this thatnotionssuchasthespeaker’s empathywith particularparticipantsandother ideaswith a strongfunctionalistflavor werecrucially implicatedin acomprehensivetheoryof reflexives.
Otherwriters have taken this point quite seriously;but ratherthanconclud-ing thatanaccountof reflexiveslies outsideof grammar, they have respondedtoit by developinga morefinely articulatedtheory. Building on considerablepre-vious work (muchof it summarizedin Koster& Reuland1991),a view that isoftenidentifiedparticularlywith thework of ReinhartandReuland(1991,1993)distinguishes“local anaphora”from “non-local” or “logophoric” anaphora.Theformerrepresentstheclassof reflexivesthatfall undera structuralconditionlikePrincipleA; non-localreflexives,on theotherhand,representaquitedistinctcat-
6
egory, with quite distinct logical andgrammaticalpropertiesfor which a ratherdifferentaccountis developed. The distinctnessof thesetwo categoriescanbeshown not only from language-internalconsiderations,but also(asmy colleagueSergey Avrutin hasargued)from theempiricaldataof child languageacquisitionandaphasicimpairment. The resultof keepingthemdistinct is a muchsharperandmorenearlyadequateformal theoryof anaphoricbinding, togetherwith atleasttheoutlinesof quitea differentareaof grammar.
Noticethat thepoint of this analysisis neitherthedenialby “formalists” thatthepointsmadeby a functionalistanalysislikeKuno’sareinterestingor relevant,nor therejectionof theideathatastructuralprincipleof anaphoricbindingis onecomponentof an autonomous,modularsyntactictheory that governsimportantregularitiesof the structureof language.Rather, what (formalist) syntacticianssuchasReinhartandReulandseein theseobservationsis evidencethat the ap-parentlyunitarycategory of “reflexives” is actuallyinternallydiverse,with somecasesfalling underonesub-theoryof grammar(thetheoryof binding,construedfairly narrowly asasetof purelystructuralprinciplesoperatingoverphrasemark-ers) and other casesfalling underanother(the theory of non-localbinding, orlogophoricreference).It is preciselybecausetheoverall category of “reflexives”involvesthe interactionof principlesfrom (at least)two quite distinct domainsof grammarthataunifiedaccountseemselusive,asFasold(1996)argues;but theright responseto thatstateof affairsis notto reject“formalism,” but ratherto workat teasingapartthefactorswhosejoint contributionsareevidentin thecomplexityof thesurfacefacts.Wewill seethatthiscaseis notatall atypical.
Within this samegeneralareaof grammar, the theoryof reference,we cannoteanotherinterestinginfluenceof functionalistwork on theoriesof a clearlyformalistcharacter. Surelyoneof theareasin whichfunctionalistviewshavebeenmostextensivelydeveloped,continuingfrom thestudiesof thePragueSchoolandearlierup throughthe presentday, is the grammaticalorganizationof sentencesto reflectnew vs. old informationandrelateddistinctions. While this literaturehasgenerallyrelied on unformalizedintuitions aboutthe informationstructureof sentencesandof the discoursesin which they occur, not even the mostrabidformalist would deny that thereis muchof greatvalueand insight to be foundthere.
Quiteexplicitly buildingonthisbackground,therelatedpositionsof DiscourseRepresentationTheory(Kamp& Reyle 1993,Kamp1981)andFile ChangeSe-mantics(Heim 1989), and more broadly the “Dynamic Semantics”movementhave madeconsiderableprogressin formalizing many of the relationsbetweendiscourseandreferencebroughtoutby functionalistaccounts.In fact,I ampretty
7
surethat the theoryof referencethat emergesfrom theseconsiderationsis justwhat is neededto under-pin the theoryof non-localor logophoricanaphorare-ferredto justabove,but I will notdevelopthatclaimhere.
I dowantto note,however, thatfurtherextensionof thenotionsof File ChangeSemanticsto encompassevenbroaderareasof theinformationalstructureof sen-tencesin a formal theoryarecurrentlyquite anactive areaof research,asillus-tratedby thework of Vallduvı (1992). Theproductivity of this particularformalexplication of core functionalistinsights,when taken togetherwith a syntacticanalysisalongstandardlines, will be evident in the analysisof Japanesepost-verbalconstructionspresentedby Kaiser(1996).
In short,then,somefunctionalistwork is in factcloseenoughto thatof yourcanonicalformalist to make dialog andreciprocalinteractionfairly straightfor-ward, andsomeareasthat have beenprimarily cultivatedby functionalistsmaywell bequitesuitablefor formalization. Let us rememberthat this is a two waystreet: after all, Kuno was led to his observationsabout“speaker empathy”innon-localbindingcasesby notingapparentinadequaciesin existing formal theo-riesof thegrammarof reflexives. Otherfunctionalistwork, suchasmuchof theliteratureon the informationstructureof sentences,implies (thoughit doesnotprovide) a kind of explication that canoften be pursuedin the developmentofformal theories,typically with someprofit in theform of abetterdevelopedsenseof theinternalstructureof thedomainof phenomenainvolved.
At the other extremeof functionalistviews we find the positionalludedtoabove which maintainsthat seriousanalysesof languagecannotin principle beformalized,or thatof linguistswhoarguethatall propertiesof grammaticalstruc-turereduceto mattersof understanding,pragmatics,communication,etc. — thatis, that thereis no distinctrealmof grammarat all. This view seemsto fly in thefaceof thefactthatthestudyof grammarhasapparentlymadequiteconsiderableprogressby assumingthatthereis somethingthereto study. I cannotreallyclaimto understandthis position,though,or why anyonewould actuallywantthingstowork out thatway, andsoI will notconsiderit further.
3 Typologyand the Functionalist Agenda
The remainderof this paperwill primarily concernthe variety of functionalismassociatedwith the studyof typology, especiallyasexemplifiedin the work ofWilliam Croft (cf. Croft 1990,1991,1995b).Amongtypologists,Croft hasbeenrathermoreexplicit thanmostaboutjust how heseesthedifferencesamongfor-
8
mal andfunctionalapproachesto language,andhis interestscomequiteclosetoonesI havebeenconcernedwith myselfat varioustimes.
“Functional/Typological” syntaxhasoften beenpresentedas the alternativeto “Formal” syntax,especiallywith respectto the discovery and characteriza-tion of universalsof language. This tradition seesitself as originating in thework of Greenberg (1963). Despitethe senseamongfunctionaltypologiststhatthey constitutethe more or lessofficial oppositionin a discussionoften domi-natedby formalists,therehasactuallybeenvery little real debate.As notedbyMatthews1993,“for mostof thepastfifteenyears,despiteoccasionaldisparage-ment from onesideor another, eachschoolhasin practicehad little reasontoreferto theother. It is worth noting,for example,that[Croft 1990]citesno workby Chomsky. Seven yearsbefore,Newmeyer, in a book on GrammaticalThe-ory whosereferencesarewideranging[. . . ] mentionedGreenberg andhisschoolonce.” [p. 45]
In general,advocatesof eachapproachto languageuniversalshave tendedtodisparagethetheother. Functionaltypologistscommonlyadoptanattitude(whichI personallyfindcompletelyunwarranted)to theeffectthatformalist(or toperson-alizethematter, “Chomskyan”) linguistswork only onEnglishandthusignoretheevidenceof a wide varietyof languages.Sociologicallycharacterizedformalists,on theotherhand,claim thattheanalysesof typologistsaresocompletelysuper-ficial asnot really to be worthy of considerationin evaluatinghypothesesaboutgrammar. Recently, it shouldbenoted,we have hadsomemoreilluminating ex-changes,suchasasthatbetweenNewmeyer1992andCroft 1995a,aswell astheMil waukeesymposiumat which the presentpaperwaspresented,wheresome-thing morein theway of mutualunderstandinghasbeensought.It still remainsto beseenhow muchof that thereis to befound,though.Piousandconciliatoryintentionsdonotconstituteresultsin themselves.
How muchdifferenceis therereally betweentypologyandformal grammar?In this connection,it is illuminating to look at a recentvolumesurveying Ap-proachesto LanguageTypology(Shibatani& Bynon1995).In additionto Croft’sGreenbergianview (entitledsomewhateclipsingly“ModernSyntacticTypology”)anda paperby Greenberg himselfon therelationbetweentypologyandchange,we find presentationsof the work of the PragueSchool,of groupsin Paris, St.Petersburg andColognewhoseindividual approachesI will not attemptto char-acterizeherebut which areeachdistinct in variouswaysfrom thatpresentedbyCroft andGreenberg.
Thelastpaperin thevolume,however, Fukui1995,approachesthetypologicalcharacterizationof a languagefrom thepointof view of straightforwardformalist
9
(“PrinciplesandParameters”)theory. On this picture,the syntacticsystemof alanguage,or at leastimportantcentralaspectsof thatsystem,areto bespecifiedin termsof valuesfor theparametersprovidedby UniversalGrammarthatdefinea limited rangeof possiblevariationin grammars.Rhetoricaside,it is hardto seehow this projectdiffersfrom thebasicgoalsof theotherconceptionsof typologyrepresentedin thisvolume,except,perhaps,in theexplicitnessof its formulations.Exploringthedimensionsof typologicalvariationisnotdifferentin principlefromexploring theparametersof variationin grammar. On theotherhand,developinga theoryof UniversalGrammardoesnot make sensewithout considerationof arangeof languages,a point which is quiteclearto Fukui andotheradherentsofthePrinciplesandParametersprogram.
If thereis a differencebetweenformalistandfunctionalistwork in typology,then,it wouldseemthatthedistinctionsareto besoughtnotsomuchin thebasicquestionsaboutthe subjectmatterasin the rangeof answersthat areproposed.And in fact, therearecleardifferencesin what countsasan analysisin the twoliteratures.As Croft explainsin somedetail in his introductorytext (Croft 1990),thefunctional-typologicalapproachseeksto find (implicational)hierarchiesin thedataof inter- andintra-linguisticvariation,andcharacterizesbasiccategoriesandtermsof linguisticanalysesby scales,gradients,andfuzzy-edgedprototypes.For-malistaccounts,in contrast,seekdiscrete,categorialanalysesandfully explicateddistinctions.Rich deductivestructureis soughtnot somuchasanendin itself asameansto reducingtheprimitivesof grammaticalvariationto a logicalminimumof genuinelyorthogonaldimensionsof variationamonglanguages.
In thenatureof things,sincethey seekto find groundingfor grammaticalphe-nomenain “language’s functionin a largercontext,” functionalistsexpectto findsubstantialinter-penetrationof phenomena.Theholisticview thateveryaspectoflanguageis permeatedby suchfactorsof meaninganduseis presentednot asanempiricalhypothesis,but rathera basicresearchstrategy. Whenever theseseemrelevant to a full accountof linguistic behavior, functionalistsfind confirmationof their view that languageis not to beexplainedby autonomous,self-containedmodules.
Formalists,in contrast,respondto suchobservationsby proposinga divisionof labor, with variouspartsof theexplanationprovidedby distinguishablecom-ponentsof linguistic knowledgewhoseworking canbedescribedindependentlyof oneanother, andwhoseinteractiongivesriseto thefull complexity of thephe-nomena.This resultsin a high degreeof modularity, which in turn supportstheautonomousformulationof variouspartsof grammar.
What is at stake hereis not just a differenceof intellectualstyle. I think it
10
actuallyreflectsa ratherfundamentaldifferencebetweentwo notionsof whatlin-guisticsis about.Theconfirmationfunctionaliststendto find for thenecessityofcontinuous,scalar, prototype-basedaccountsis relatedto thefactthattheobjectofinquiry in (much)functionalistwork is really ratherdifferentfrom thatin (much)formalistresearch.If whatthelinguist is interestedin is thecapacityof thehumanlanguagefaculty, asmostcurrentschoolsof formalistsyntaxwouldmaintain,thendataaboutusageandfrequency — or eventheissueof whetheragivenpossibilityis everactuallyinstantiatedin any language— maybeof atmostrathermarginalinterest.On theotherhand,if (alongwith muchof thefunctionalistcommunity)whatyouseekis aunifiedaccountof theactivityof humanlanguageusers— whatspeakersdo with language— thenall of thesedomainsthat go beyond just theconsiderationof whatthehumanlanguagefaculty is seemcrucial.
I have arguedelsewhere(Anderson1992, chap. 13) that the scopeof thelanguagefaculty cannotbe derivedeven from an exhaustive enumerationof thepropertiesof existing languages,becausethosecontingentfactsresult from theinteractionof the languagefaculty with a varietyof otherfactors,including themechanismsof historicalchange.To seethat what is naturalcannotbe limitedto whatoccursin nature,considertherangeof systemswe find in spontaneouslydevelopedlanguagegames,assurveyed by Bagemihl(1988). Theseobviouslyincludemany sortsof transpositionsandmanipulationsof phonologicalform thatcorrespondto no occurringphonologicalor morphologicalrule of any language,but which seemnonethelessto fall within thecapacitiesof languageusers,oncedevelopedfor thespecialpurposessuchsystemsserve.
I would arguethat the reasonfor the non-occurrenceof many of thesepro-cessesin naturallanguagesis simply that thereis no plausiblepathof linguis-tic developmentwhich, startingfrom any attestedsystemandproceedingby thepossiblemechanismsof phonetic,phonologicalandmorphologicalchange,couldgive riseto them. It is preciselywhenthecontentof thesystemis divorcedfromsuchnaturaldevelopmentthat its full potentialrangecanbeexplored,andwhenthathappens,we seereasonto believe that theunderlyingfaculty is ratherricherthanwemighthaveimaginedevenonthebasisof themostcomprehensivesurveyof actual,observablelanguages.Thisdoesnotmeanthatempiricalobservationisirrelevant to anunderstandingof the languagefaculty, but it doesmeanthat it isnot sufficient, andin factobservationsaboutpreferences,tendencies,andwhichof a rangeof structuralpossibilitiesspeakerswill tendto usein a givensituationare largely irrelevantto anunderstandingof whatthosepossibilitiesare.
On theotherhand,if you wantto capturethefull reality of linguistic activity(asopposedtocapacity),thenissuesof attestation,frequency, tendencies,marked-
It seemsto methatin its attemptto treat“Language”in themostcomprehen-sive senseasa unitaryobjectof study, thefunctionalistview systematicallycon-foundstheseeffectsin a way thatresultsin a considerableamountof intellectualconfusion.It is far from obviousthat theintersectionof all thesefactorsactuallyconstitutesa coherentobjectof studyin its own right. In the naturalworld, forexample,thewind is certainlya phenomenonwe canidentify, but would it makesenseto try to developa unitaryscienceof wind? Surely“wind” resultsfrom theinteractionof a wide rangeof climatic, geographic,atmosphericandother fac-tors,andtheway to understandit is astheproductof their interaction— not asa unitary objectof studyon its own. Similarly, I seeno reasonto abandontheassumptionthat an extensively modularapproachis the bestway to attacktheproblemof findingtheorderandcoherencein languagethroughscientificinquiry.
4 A Styleof Functionalist Ar gument
Someof the issuesseparatingformalist andfunctionalistmethodologiescanbeillustratedin relationto thefollowing basicprincipleof epistemology, originallyattributedto WalterReuther:
(1) “If it looks like a duck,walks like a duck,quackslike a duck, thenit is aduck.”
We might approachthis asa strategy for the scientificinvestigationof someimportantconceptin grammar. Indeed,it is fairly routineto developsomeseriesof “tests” to whichwecansubjectlinguisticobjectsto determinetheirstatuswithrespectto significantcategories:for instance,to determinewhetheror not some-thing is “really” anobjectin a givensentence,we mightaskwhetherit is markedwith Accusativecase,canbecomethesubjectof acorrespondingpassive,etc.
Croft 1991developsthis sort of strategy in somedetail in connectionwithbasicissuessuchasmembershipin major lexical categories,thegrammaticalre-lationsborneby argumentswithin a clause,andothers.His strategy is to identifyasmany propertiesaspossiblethatareassociatedwith thecategory in question,
12
andtranslateeachof theseinto sucha test. Therangeof such“behavioral tests”thatcanbebroughtto bearona particularissuecangetquitelarge,asin thecaseof thepropertiesof subjectsexploredin Keenan1976(a work which canbeseenastheultimateembodimentof thisapproach).
As only a slight caricature,I think that if Croft were to apply principle (1)abovein thisway, hewoulddiscover(afteraseriousconsiderationof known factsconcerningducks)thattherearehardlyany true,full-fledgedducks.Somepoten-tial duckspassmoreof theteststhanothers— somefew mayevenpassall of thetests— but mostfail oneor moreof theteststhatwecoulddevisefor duck-hood.His responsewould probablybeto suggestthatthenotionof a “duck” is actuallya kind of prototype,to which someobjectsapproximatemoreor lesswell. Thenotionis abit fuzzyaroundtheedges,with somecentralexemplarsandanumberof nearmissesaswell, of course,asagreatmany clearnon-ducks.
Formalistsaregenerallynot contentto settlefor prototypeducks,nearducks,etc.By andlarge,they wanttheworld to beorganizedinto theducksandthenon-ducks,period. In my own contribution (Anderson1976)to thevolume(Li 1976)in which Keenan’s paperappeared,I tried to deal with the fact that supposedpropertiesof subjects,whentranslatedinto testsandappliedto ergativelanguages(to which a numberof otherpapersin the presentvolumesarespecificallyde-voted),turn up a surprisinglyrobustsetof somewhat-but-not-really-quiteducks.My strategy wasbasicallyto take somepotentialphenomenaascriterial, whilerelegatingothers(particularlymattersof overtmorphologicalform) to aseparate,non-criterialstatus.Croft (1991)takesmeto taskfor makingsuchanapparentlyarbitraryselection,apparentlymotivatedonly by my desireto comeup with acompletelyconsistentway to tell theducksfrom thenon-ducks.
If thathadindeedbeenall thatwasat work in my analysis,Croft’s criticismwould have beenwell warranted.Actually, though,therewasa bit moregoingon. Indeed,the constructionandselectionof someparticularsetof testsfor acategory, while quiteastandardactivity in thefield, is pointlessuntil sometheoryis presentedin termsof which the criterial natureof the testsmakessenseandfindsits basis.
Insteadof consideringprinciple(1)aboveassynthetic,asanempiricalstrategyfor picking out ducksin nature,let us take thealternative view that it is analytic:that is, a Theoryof duckswould be somesort of systemof propositionswithinwhich variouspropertiesand entities(including ducks)appear. Insofar as thesystemcorrectlyreconstructsourintuitionsaboutducks,wecansaythatit definesaduckassomethingthatdisplaysthesepropertiesandsatisfiesthesepropositions.In otherwords,insteadof assumingthatweknow all aboutducksin advance,and
13
aretrying to give aneffective field guide,let ussupposethatwhatwe aretryingto do is to reconstructthe contentof beinga duck in scientific terms. Insteadof beingcontentwith the observation that somethingsaremoreduck-like thanothers,wewouldliketogoonto developamoreexplicit accountof thepotentiallyindependentdimensionsof duckiness.
ThatwaswhatI wastrying to doin thepaperreferredto above,alonglinesthatarefairly standardin the formalist literature. I startedfrom thepremisethat thenotionof “subject” in pre-systematicdiscussionwaspotentiallya heterogeneousone. Fromthepoint of view of thesyntax,it seemedplausiblethat thenotionofbeingasubjectcouldbedefinedin termsof positionwithin aphrasemarker, onthepremisethatthestructureandinter-relationsof phrasemarkersaretheessenceofsyntax.Again from thepointof view of thesyntax,structure-sensitiveoperationsthat mapphrasemarkers into otherphrasemarkersought to treatelementsin aconsistentwayon thebasisof theirpositionin aphrasemarkeralone.To identifyanotionof “subject”within thesyntax,then,whatonewantsto do is to find asetof suchoperations,andtreatthemascriterial for thesyntacticnotionin question.In termsof then-currenttheory, teststhat seemedto correspondto this criterionincludedthelocationof PRO in infinitival constructions(in thevocabulary of thetime, thetargetof “ EQUI-NP deletion”),thebindingof reflexives,andthenotionof whatpositionscountas“parallel” in theformationof coordinateconstructions.
WhenI appliedtheseteststo a variety of “ergative” languages,I found thatthey yielded resultsthat were as consistentin thoselanguagesas they were inaccusative languageswith respectto whenanNP constitutesa subject;andfur-thermore,thesameNP’sturnoutto bethesubjects,by andlarge,in sentencesthataretranslationequivalentsbetweenthetwo sortsof languages.I took this to con-firm thenotionthat thereis a consistentnotionof subject,internalto thesyntax,which is remarkablyinvariantacrosslanguages.
Now of course,beingasyntacticsubjectis only partof themassively complexnotionof “subject” thathasdevelopedin talk aboutlanguageover thepastcou-ple of thousandyears,andit wasactuallythis complex andheterogeneousnaturethatwasKeenan’smajorpoint. Theveryexistenceof ergative languagesmakesitclearthat thesyntacticsubjectwill oftenfail to meetour morphologicalexpecta-tionsabouttherelationof thesubjectto theformsof words(thatsubjectsshouldappearconsistentlyin thenominative case,theverbshouldagreeconsistentlyorexclusively with them,etc.). Indeed,the factsof Dyirbal make it clearthat thesyntacticsubjectwill not alwaysmeetour expectationsaboutthe semanticsofsubjects(thatthey shouldbeagentswherepossible,etc,)either.
14
But to identify a syntacticnotion of subjectis not at all to denigratetheseothernotions,or relegatethemto second-classstatus:it just meansthatwe haveto explicateotherdimensionsof thecomplex concept“subject” in otherwaysandwithin othersub-theoriesof grammar, suchasthetheoriesof how syntacticorga-nizationis relatedon onehandto semanticargumentstructure,andon anothertothecategoriesof overtmorphology. If someof theseothertheoriesareunusuallysimpleandconsistent,it might turn out that syntacticsubject-hoodwould be aconsistentpredictorof everythingelse,but of courseKeenanhasalreadyshownthatthingsarenot thatsimple.
A particularlystrongtendency thathasa lot of traditionbehindit is to assumethat the theorythat relatessyntacticstructureandmorphologicalclasswill turnout to beespeciallysimple,suchthattheinflectionalcategoriesof wordform willturn out to be essentiallyisomorphicwith the categoriesof the configurationalsyntax.Perhapsthatexpectationarosefrom thefact that for muchof thehistoryof grammaticaldiscussion,therehasnot reallybeenany distincttheoryof config-urationalsyntax,andwhatwentby thenameof syntaxwasmostly just “appliedmorphology.” But wehavenotbeenin thatstatefor thepastforty yearsor so,anda seriouslook at therelationbetweenmorphologicalandsyntacticform turnsupconsiderablymorearbitrarinessandcomplexity thanweareusedto expecting,asI tried to show a few yearsagoin Anderson1991. To theextent that is true, theinteractionsof modulesof grammarmaybefar from trivial.
On the otherhand,thereis alsono needto acceptthe conclusionthat whenit comesto the distribution of subjectpropertiesin individual languages,everydog is from a different village: that is, that the thirty-odd propertiesof (pre-systematic)subjectsdiscussedby Keenanarein principledistributedcompletelyindependentlyof oneanother. In fact, thesemayclusterin coherentways,sincea numberof propertiesmayreflecta commonorganizationwithin somedomain,oneof which oughtto bethehierarchicalorganizationof phrasemarkersandtheoperationsthataffect these,if syntaxreallyhasthecharacterwe think it has..
The methodologicaloppositionhereis the following: to judge from muchfunctionalistwriting, the pre-systematiccategoriesof traditional grammararewidely takento havea life of theirown — thereis a tendency to assumethattheyareunitaryandimmutable,andthatour taskis to identify them. Wherewe findthatour theoriesdonotprovideacompleteandcategoricalreconstructionof sucha notionas“subject,” we might concludethat it is actuallyfuzzy, prototype-like,etc.But why shouldweexpectanotionlike “subject,” with all of theaccretionofobservationsthat it hasacquiredover theyears,to correspondto a unitary theo-reticalcategory? Our goal in doing linguistics,afterall, is to reconstructnot the
15
specificnotionsof traditionalgrammar, but ratherthe intuitionsandinsightsthatunderlaythem.
Thealternative is to developa theorythathassomereasonablynaturalcleav-ages,suchas“structuralsyntax”vs. “morphologicalform,” “discourserole,” “se-manticrole,” etc. In eachsub-area,wedevelopacoherenttheorythatreconstructsto theextentpossibletheinsightsof traditionalgrammar(sometimesfinding thatapparentgeneralizationswerein factillusory or epiphenomenal),aswell asa the-ory of how principlesfrom thevariousdomainsinteract.Theresultis a view onwhich thesyntacticducksmayor maynotbemorphologicalducksaswell, but inwhich we have a numberof individually coherentsensesof “duck vs. non-duck”ratherthana single,monolithicbut ratherfuzzynotionof “duck,” toutcourt.
ThereasonI have rehearsedthesemattersis not that I think it is importanttodefendapaperI wrotetwentyyearsago— onewhosespecificassumptionsaboutsyntaxare probablyindefensiblewithin today’s theories. Rather, it is becauseI think thereis a rathergeneralpatternof argumentationto be discernedin thefunctional/typologicalcommunity’s reactionto thatpaperandotherslike it, andananalysisof thatreactionis importantif onewantsto understandhow formalistsrespondto whatthey readin functionalistwork.
The(functionalist)argumentsin questionareintendedto establishtheneces-sity of anappealto extra-grammaticalfactorsin analyzinggrammaticalstructure— that is, the impossibility of the kind of autonomousanalysissoughtby mostformalistsyntacticians.Now whenthey look for thebasicobjectsof investigationin linguistics,functionaliststendto seeoverall classesof constructions,oftende-finedby semantic/pragmatic/discoursefactors.But we have to askwhatwarrantwe have for assumingthat the descriptive taxonomyof previous generationsofgrammarianshassucceededin identifyingunitaryphenomena.How do we knowthatanotionlike“subject” is conceptuallyhomogeneous,or eventhatthereis anysenseto comparing“passive” constructionsacrosslanguages,etc.?
What happens,however, is that somevery broadly conceived term suchas“subject,” “passive,” “agreement,” etc. from traditionalgrammar, togetherwith allof its pre-systematicallyassociatedbaggage,is presentedfor analysis.Whenallof therelatedfactsareconsidered,it is usuallypossibleto show thatsomeof themareintrinsicallygroundedin factorsoutsideof thesyntacticcomputationalsystem.Therefore,it is suggested,anautonomoussyntacticanalysisof this phenomenonis impossible(or atbest,misguided).
The problemwith this argumentis not with the supposedextra-grammaticalfactorsthemselves,but ratherwith theassumptionthatthetheoryoughtnecessar-ily to treatthe termor constructionin questionasa singlehomogeneousobject.
16
Whenwe look moreclosely, it usuallyturnsout that the singletraditionaltermconcealsan identificationof phenomenafrom severalareasof grammar, eachofwhichindividually hasaperfectlycoherentanalysiswithin someappropriatesub-discipline in the studyof language,but whereno singlecomponentsuffices toexplicateeverything.Theformalistresponseto suchanargumentis thatas,say, asyntactician,heor sheis concernedwith whatis syntactic,but therestis someoneelse’sbusiness,andthereis nothingin thesyntacticfactsthathavebeenpresentedthatnecessarilycompromisestheadequacy of a formalsyntacticaccount.
I think that is what is going on in the caseof the discussionof notionsof“subject” in ergative vs. accusative languages:the strictly syntacticphenomenaacrosslanguagesconduceto a ratherhomogeneousandcoherentaccount,thoughtheconnectionbetweentheseandotherphenomenathathave beenlinkedto thetraditional term “subject” is much more varied. We canseethe samelogic inwork that respondsto Kuno’s observationsaboutreflexivesby developingdis-tinct theoriesof local binding,construedrathernarrowly in termsof the syntax,andnon-localbinding, a ratherdifferentrelationthat is groundedin the formalstructureof discourserepresentations.
Another instructive exampleis provided by an argumentCroft (1995a:501)offers“againsttheself-containednessof syntax.”:
Babungois aGrassfieldsBantulanguage(Schaub1985).Its nounclasssystemis typical of Bantu languages,in that nounsfall intoa rangeof nounclasses,andthe nounclassesarequite arbitrary intheirmembership.Thatis, thereis nowayonecouldcompletelypre-dict nounclassmembershiponsemanticprinciples,althoughSchaubnotessomecorrelations,e.g. class1/2 with humans(1985:174;thepairednumbersreferto singular/pluralforms).Theseagreementpat-ternsindicatethatnounclassesarenot only arbitrarybut participatein a systemof syntacticrules.Modifiersagreewith their headnounsin class,andverbsagreewith their coreargumentsin thesameway.Thereis oneexceptionto thisgeneralization:anaphoricpronounsthatreferto humans,regardlessof thenounclassof thehumannoun,takeclass1/2agreement(Schaub1985:193).Anaphoricpronounsthatre-fer tonon-humansagreein thenounclassof theirnominalantecedent.
For Croft, theevidentinvolvementof semanticfactorsaswell asformalnounclassin thechoiceof pronounsconstitutesclearevidencethat“thesyntaxof agree-mentin Babungo” cannotbeanalyzedin a self-containedway. But I think mostformal syntacticianswould deny that in Croft’s intendedsensethereis any such
17
thing as“the syntaxof agreementin Babungo” asa unitaryphenomenon.On theonehand,thereareagreementphenomenainvolving (on oneformulation)somesortof SPEC-headfeaturecheckingor assignmentwithin theclause,andthissys-tem is responsiblefor theagreementin nounclassbetweenmodifiersandheadsandbetweenverbsandtheir argument. We have no reasonto believe thesearemattersof anything but sentenceform. On theotherhand,thereis thematterofthe selectionof an appropriatepronounto refer to an antecedentthat hasbeenmentionedelsewherein a discourse.This is surelynot at all a matterof sentencesyntax,however, andthereis no reasonto treat it assuch,despitethe fact thatthenounclasssystemwhich operateswithin sentencesis oneof the factorsthatcontributesto pronounchoice.
Considerthefollowing sentence:
(2) (My cat)Dent-de-lionis still hungry, evenafterit/she/hehasfinishedall ofits/her/hisbowl of Skrunkies.
Thechoiceof a pronounheredependson people’s views aboutthe sentienceoftheir pets,knowledgeof their secondarysexual characteristics,the existenceofNounsin otherlanguagesbelongingto arbitrarygenderclasses,etc. Would any-onearguethatthesefactorscompromisetheclaim thatsubject/verbagreementinEnglishis a formal process?Thereare,it is true,many unresolvedpuzzlesin thesyntacticanalysisof Englishagreement,but thechoiceof pronounsfor discoursereferentsis surely not amongthem, becauseit seemsclear that entirely differ-ent componentsof linguistic knowledgeareat work here. Indeed,a theorythatfailed to distinguishtwo quitedistinctsensesof agreementherecouldbesaidtobeinadequateon thatbasis.
This point is mademoreconcretein a recentpaperby FarkasandZec(1995)whichdealswith factsin Roumanianthatareentirelyparallelto thoseof Babungoalludedto by Croft, andtherebyshows how a formalistaccountof “agreement”mightproceed.FarkasandZecassumethatthesyntaxinvolvesrulesof agreementin termsof a setof morphosyntacticallyrelevantfeatures(SG, PL, I, II, III, Mas-culine,Feminine)whicharedistributedby someprocess(es)of featureconcordsothatmodifiersagreewith their headsandverbswith their subjects.They alsoas-sumethatan independentlevel of DiscourseRepresentationStructure(alongthelines of Kamp1981andHeim 1989)identifiesdiscoursereferentsin termsof adistinctsetof features(including [atomic], [group], [ � Participant],[ � Speaker],[male], [female]). A systemof rules is developedthat specifiesthe correspon-dencebetweentheDRSpropertiesassociatedwith a referentandthe grammati-cal featuresof thecorrespondingexpression(includingdiscourse-referentialpro-
18
nouns). Theserules accommodatepuzzleswith grammaticalgenderthat havebeenpointedout by Corbett(1990),suchasthosemotivating“resolutionrules”.Theresultis at leasta substantialstarton a full, formal reconstructionof agree-mentphenomena,involving the independentpropertiesof at leasttwo domains(morphosyntacticstructureanddiscourserepresentation)andthe systematicitiesin their inter-connections.
It is only by lumpingtogetherall of theheterogeneousphenomenathatareas-sociatedwith thenotionof “agreement”thattheBabungofactsappearto compro-misetheautonomyof syntacticanalysis.But thegoalof a formally seriousanaly-sisis preciselyto disentangletherelatedbut fundamentallyindependentstrandsofa superficiallycomplex phenomenon.As a result,argumentsin this style,whichseemto reflectanessentiallyincompleteanalyticunderstandingof the structureof thefacts,have little directforcefor thebeliever in formally serious,essentiallymodularanalyses.This is thesortof discussionthatDavid Perlmutterusedto call“helicopterlinguistics”: youfly overthephenomenaafew times,andassumethattheview thatyougot from theheightsrepresentstheendof thestory.
Whenwelook at thefunctional/typologicalliterature,unfortunately, thisstyleof argumentseemsall too often to underliethediscussionsthatpurportto showthe necessityof including extra-grammaticalfactorsin the analysis,or the non-categorial, fuzzy-edgednatureof somegrammaticalterm or category. This iscertainly true for the discussionof grammaticalrelationssuchas “subject” or“object,” andthesameconfusionextendsdirectly to mostfunctionaltypologicaldiscussionsof ergativity. Ergativity, when examinedclosely, turnsout to be acategorialparameter— or rather, severalsuchparameters,primarily affectingtherelationbetweensyntacticstructureandthematicor semanticstructure,ontheonehand,andmorphologicalstructureon theother. In any case,“ergativity” asauni-tary phenomenonis a matterwhosetypologicalsignificanceis unclear. What isgenerallyreferrredto in thetypologicalliterature(incorrectly, it appears)as‘Syn-tactic’ ergativity is (roughly) a questionof whetheragentsor patientstypicallyappearasthesubjectsof transitive Verbs.Morphologicalergativity is a propertyof particularrules,dependingonwhichof two possibletypesof referenceto syn-tacticstructureis madeby agivenrule(seeAnderson1992for somediscussionofthemechanismsinvolved). Theremayor maynot bea relevantparameterwithinthesyntacticsystemsensustricto(seeMarantz1996for asurvey of viewsonerga-tivity in recentformalsyntax).Eachof thesenotionsis quitecategorical,onceweareclearabouttheproperlocuswithin thegrammaratwhichto raisethequestion.
Exactlysimilarremarkscouldbemadeabouttheclaimof Hopper& Thompson1980thattransitivity is agradient,notacategorialnotion.If youidentify all of the
19
phenomenathatclusteraroundverbalvalenceandargumentstructureasa singleconstruct“Transitivity,” it is not surprisingthat the reconstructionof transitiv-ity turnsout to be thoroughlyheterogeneous— indeed,whatwould beamazingwould beif thingsturnedout any otherway. But that in itself doesnot constituteanargumentagainstthecoherenceof a straightforwardly syntactic,if ratherlessambitious,conceptof transitivity, namelythe presencevs. absenceof an objectargumentbearingaspecifiablestructuralrelationto ahead.Theotherfactorsthatcommonly(butnotuniversally)clusteraroundthisstructuralnotionhaveperfectlycoherentanalysesof theirown thatdonot impugnthesimplicity andgeneralityofstructuraltransitivity.
A standardrhetoricalfigurein thefunctional-typologicalliteratureis to iden-tify a phenomenon,and then provide an explanationfor it that crucially relieson appealsto extra-grammaticalfactorssuchasdiscoursetopicality, metaphorsgroundedin thespeechact situation,etc. Unfortunately, the objectsof suchex-planationsare sometimesepiphenomena,the result of spuriousgeneralizationswhich collapseon detailedanalysis.An exampleof this is the apparentcorrela-tion, in some“split ergative” systems,betweencasemarkingandverbalaspect:ergative markingis sometimesfound in associationwith perfective verbalforms(or their reflexes,suchasHindi pasttenseforms) in a singlelanguagewhereac-cusativemarkingis usedwith imperfectiveverbalforms.Theapparentconnectionbetweencasesystemsandaspecthasbeensomethingof a stapleexplanandumintypologicaldiscussion(seeDelancey 1981,Croft 1995b,pp. 120f.).
In fact, however, asdiscussedin Anderson1992(pp. 354ff.), this apparentcorrelationis probablyaccidental.It happensthat two commonsourcesfor thehistoricaldevelopmentof perfectaspectformsare(a)passives,and(b) possessiveconstructions,eitherof which canleadto a constructionin which the (agentive)subjectappearsin anobliquecase(which maybereanalyzedasanergative). Ontheotherhand,onesourceof innovative imperfective forms is an“object demo-tion” construction,which may result in the markingof objectswith an obliquecasethat is re-interpretedasanaccusative in anoriginally ergative languagelikecommonKartvelian. Thesetwo developmentsarequiteunrelatedto oneanother,but happento leadto similaroutcomesin termsof thedistributionof casemarkingacrossaspectuallydifferentforms.Thedetailsarecomplex, but oncetheindivid-ualcasesareanalyzed(ratherthanbeinglumpedtogether),it appearsthatthereisnomoreto explain in thiscasethanin othercasesof accidentalconvergence,suchaslexical homophony.
analysisor the gratuitousacceptanceof traditionalcategoriescould be attestedfrom the literatureon “NP-identificationphenomena,” a notion which conflatesthedistributionof null pronominalsubjectsof infinitives,sharedNounPhrasesinconjoinedexpressions,NP-pronouncoreference,identificationof discourserefer-ents,andmany otherareasof grammar.
Theoppositeof argumentsfrom over-generalizationis theargumentfrom ex-cessive particularization,which is also prominentin the functional typologicalliterature. With respectto ergativity, it hasbecomecommonnow to point outthatwhole languagescannotmeaningfullybecharacterizedas“ergative” or “ac-cusative,” sincewe generallyfind somephenomenawithin a givenlanguagethatcouldberegardedasorganizedonanergativepattern,andothersin thesamelan-guagethatareequallyaccusative. Theresponseto thisof work suchasCroft 1990is to suggestthat suchtypologicalparametersapply not to languagesbut to in-dividual rules. But this move goestoo far in the other direction: in fact, thephenomenawithin any givenlanguageshow somesignificantclusteringrelations,correspondingto thenaturaldivisionsof thegrammar. Therulesof inflectionre-lating syntacticandmorphologicalcategoriesmayindeedvary onefrom anotherin this way, but the rulesof the syntaxitself appearto be foundedon a single,coherentnotionof structure.
In thiscase,asin all of theothers,thecrucialstepto aninsightfulanalysisis tofind theright granularity, themostappropriatedecompositionof a complex phe-nomenoninto individually coherentpieces.And of course,a centralrole in thatprocessis playedby thechoiceof anappropriateformalism,sinceit is reallyonlyaftera potentialanalysishasbeenlaid out formally thatits componentsarereallyavailablefor inspection.Ultimately, linguistsof all stripeshave to provide a for-mally explicit accountof theiranalyses,asCroft (amongsomeotherfunctionalistwriters) clearly recognizes.But despitethis commitment,it is remarkablehowrarelya fully explicit analysisof the grammarof relevant languagesis providedto underpinfunctionalistdiscussion.Sometimestheabsenceof suchananalysisis attributedto disagreementsaboutwhatthe“right” theoryof grammarmightbe,asif suchdissensionobviateda detailedaccount.In fact,however, the longerafull explicationispostponed,asit tendstobein functionalistaccounts,andpresys-tematicdescriptivecategoriesareallowedto standin for thefully explicit account,themorelikely it is that the discussionwill go astrayon the basisof suggestivecorrelationsthatdonotultimatelycorrespondto primitivesof linguisticstructure.
21
5 Conclusions
Theway to understandtheoverall grammarof a language,then,is not to assumethat thereis someunitary setof maximally inclusive andrathermonolithic an-alytic categories,perhapsthoseof traditionalgrammar;nor yet to assumethatthereareno categoriesat all, only particularfacts,andthat all factsareequallycontingentandmutually independent.It seemspreferableto me,rather, to adopta ‘modular’ point of view, andaskaboutthe categoriesandtypesthat arerele-vant to anunderstandingof what is orderlyandcoherentin eachspecificareaofgrammar. This entails,for example,analyzingthe syntaxon the basisof syn-tactic phenomena,while construingthe morphologyas providing hints but notunambiguousarguments;andthenanalyzingtheprocessesof word formationinlanguageandthewaysin which syntacticinformationis used(or disregarded)intheoperationof theseprocesses;andthenproceedingto comparableaccountsofsemantics,pragmatics,discoursestructure,etc.
With respectto thestudyof linguistic typology, oftenthoughtto bepeculiarlytheprovinceof functionalistinterest,I donotpersonallybelievethereis any inter-estingdistinctionbetween“doing typology” and“doing theory.” Typologistsseekto find implicationsamonglinguistic phenomena,but this is not basicallydiffer-ent from the theoretician’s desireto elucidatethe internaldeductive structureofthe theory. Typologistsoughtnot to be contentwith merelyobservingthat onegrammaticalphenomenonis correlatedwith another;but oncethey begin to askwhat it is aboutthe natureof Languagethat leadsto the observed connections,they arelookingfor thesamefundamentalprinciplesof grammarthey wouldseekif they thoughtof themselvesas“doing Theory.”
Typologistsdo indeedpay greatattentionto the diversity of phenomenatobe found in the world’s languages,and not only to the complexity internal toindividual languages;but thesamecansurelybesaidfor any moderntheoreticianwho wantsa generaltheory. Indeed,muchformalist researchin syntactictheorytodaygoespreciselyby the nameof Comparative Syntax,andseeksto delimitexactlythefull rangeof variationfoundacrosslanguageswith respectto syntacticforms. If a typology is anexhaustive characterizationof this rangeof variation,it is just asmuchanobjectof desirefor formalistgrammariansasfor functionaltypologists.
I am afraid I have goneon at rathergreatlengthwith a somewhat arid dis-cussionof methodology, ratherthanaspicy recitationof exotic factsandtheirex-planation.But whenoneaskswhat formalistslearnfrom functionalistwork, themainpartof theansweris likely to be: they learnthatsomefurtherdecomposition
22
is oftenwarrantedin factualdomainswhichhadtraditionallybeenpresumedto belogically unitary, or elsethattheremaywell besomedeeperprincipleunderlyinga setof superficiallydiversephenomena.No onehasany reasonto doubt thatamongthefactsrelevantto languageandlinguisticbehavior in thebroadestsensearemany thatarequiteexternalto grammar. But while thefunctionalistresponseis to embracetheseasevidencefor the impossibilityof anadequateformal syn-tax,theformalistaskswherethenaturaljointswithin andamongthefactslie, suchthattheir full complexity canbeseento resultfrom theinteractionof individuallysimplesystemsbasedon broadlyexplanatoryprinciples.Both of thesereactions,of course,constituteresearchstrategiesratherthanempiricallyfalsifiablepropo-sitionsaboutwhat theworld is really like. But I think themodularline, with itsassumptionof individually autonomoussystems,hasprovensufficiently produc-tiveto bewell worthpursuing,andthatstrategy restsessentiallyonthekind of fullandcompleteexplicationthatis providedby formalization.And that,perhaps,iswhatformalistslearnfrom functionalistsin syntax.
23
References
Anderson,StephenR. 1976. On theNotion of Subjectin Ergative Languages,inLi 1976,2–23.
Hopper, Paul J.& Thompson,SandraA. 1980.Transitivity in GrammarandDis-course.Language56, 251–299.
Kaiser, Lizanne 1996. Representingthe Structure-DiscourseIconicity of theJapanesePost-VerbalConstruction.[SymposiumonFunctionalismandFor-malismin Linguistics,Mil waukee,WI, 20April].
Kamp,Hans1981. A Theoryof Truth andSemanticRepresentation,in J. Groe-nendijk, T. Janssen& M. Stokhof (eds),Formal Methodsin the StudyofLanguage: Proceedingsof theThird AmsterdamColloquium. Amsterdam,277–322.
Kuno, Susumu& Kaburaki, E. 1977. EmpathyandSyntax. Linguistic Inquiry8, 627–672.
Li, C. N. (ed.)1976.SubjectandTopic. New York: AcademicPress.
Lightfoot, David 1990. Old HeadsandNew Heads,in J. Mascaro & M. Nes-por (eds),Grammarin Progress: GLOW Essaysfor Henkvan Riemsdijk.Dordrecht:Foris,317–322.
