Light Verbs and θ-Marking Author(s): Jane Grimshaw and Armin Mester Source: Linguistic Inquiry, Vol. 19, No. 2 (Spring, 1988), pp. 205-232 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4178587 . Accessed: 03/12/2013 11:00 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. . The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry. http://www.jstor.org This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions Jane Grimshaw Armin Mester Light Verbs and 0-Marking This is a study of 0-markingwith the Japanese Verb suru, which has the apparently peculiar property of allowing the head of its direct object to 0-markargumentsat the clause level. Ourgeneralgoal in this articleis to explore the predicate-argument complex associated with suru and to show that its propertiescan be derivedfrom the interaction of complex predicate formationwith a particulartheory of argumentstructurerepresentation. Suru is thematicallyincomplete or "light" in the sense of Jespersen (1954) and Cattell (1984).1It subcategorizesand case-marksa direct object NP, without assigning it a 0-role. 0-markingin the surucomplex is a functionof a process of complex predicate formation, which we call ArgumentTransfer.The nominal 0-markertransferssome or all of its argumentsto the argumentstructureof the light Verb suru. As a result, both suru and the head of the object NP can act as 0-markers,each with its own 0-marking domain. The Noun "lends" argumentsto suru, turningsuru into a 0-markerand remainingan impoverished0-markeritself. Thus, Transferdoes not simply mergethe two predicates and their argumentstructures into a single unit, in the way that complex predicate formationis usually conceived. Ideas using complex predicate formationof various types have been developed for related English phenomena by Cattell (1984), Higgins (1974), and Jackendoff(1974). The behavior of Transferreflects certainpropertiesof argumentstructures,which are notjust lists of arguments,but form structuredrepresentations,as proposedin Grimshaw (to appear).In additionto the now familiarinternal/externaldistinctionof Williams (1981), developed in work by Zubizarreta(1985) and Levin and Rappaport(1986), we will arguethat internalargumentsare not homogeneous,but also participatein the structural organizationof argumentstructure. Whereas Nouns and Verbs generally have This researchwas supportedby the InformationScience Division of the National Science Foundation throughgrantsIST-8120403and IST-8420073to BrandeisUniversity. Additionalsupportwas providedby a BernsteinFacultyFellowshipfromBrandeisUniversityto Grimshaw.We wouldespeciallylike to acknowledge the aid of Junko Ito, who helped throughoutwith the constructionof examples and participatedin many productive discussions of Japanese grammar.Two LinguisticInquiryreviewers, Naoki Fukui, Jacqueline Gueron, Alan Prince, Tomas Riad, and MitsuakiYoneyama gave comments that led to many substantial improvements.We would also like to thankaudiencesat BrandeisUniversity,the Universityof Connecticut, and the Universityof Texas at Austin, where variouspartsof the materialwere presented. ' Discussing"light" Verbs, Jespersen(1954,VI, 117-118)commentson "the generaltendencyof ModE to place an insignificantverb, to whichthe marksof personandtense are attached,beforethe reallyimportant idea ...". Linguistic Inquiry, Volume 19, Number 2, Spring 1988 205-232 C 1988 by The Massachusetts Institute of Technology This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 205 206 JANE GRIMSHAW AND ARMIN MESTER similar argumentstructurerepresentations,Nouns do not have subject arguments.Instead, the relevantpositionin theirargumentstructureis lexically suppressedor satisfied and cannot be responsiblefor 0-markingan argument,a hypothesis suggested in Zubizarreta (1985)and developed furtherin Grimshaw(1986;to appear). We will show that the proposedrepresentationof argumentstructure,togetherwith the theory of Transfer, predicts the rather intricate pattern of argumentdistribution within the suru complex. 1. 0-Opaqueand 0-TransparentNPs 0-markingobeys strict locality conditionsand is basically restrictedto sisters (Chomsky (1981)).An argumentNP fills a position in the argumentstructureof its governingVerb, and no element inside the NP interactsthematicallywith any element outside it. In this sense, argumentNPs are opaque to 0-marking.An NP can be an argumentof a Verb, but it cannot containan argumentof a Verb. As a result, the 0-markingin (la) is possible, but that in (lb) is not: a Verb cannot assign a 0-role into an NP. Moreover, the head of an NP does not assign a 0-role outside its maximalprojection, with the result that the 0-markingin (ic) is not allowed. NP is generallyopaqueto 0-markingin both directions, presumablybecause it is opaque to governmentin both directions. (1) a. [V NP]vP b. [V ... . NP .. C. [... NP .. .[N]NP t I" II 11 .]NP]VP ... --. NP ... ]VP t Japanese has a productiveconstructioninvolving what appearsto be a case of the 0-markingin (lc). It is illustratedin (2a), where aiseki 'table-sharing',the head of the direct object NP, seems to be 0-markingJohn and Bill.2 (2) a. John-wa Bill-to AISEKI-o shita. John-TopBill-withtable-sharing-Accsuru-Past 'John shareda table with Bill.' b. John-wa Bill-to AISEKI shita. John-TopBill-withtable-sharingsuru-Past The behavior of suru in Noun incorporation(as in (2b), where aiseki is not casemarked) has been extensively discussed in Japanese linguistics in Hasegawa (1979), Inoue (1976), Kageyama(1976-77; 1982),Kuroda(1965), Poser (1980), and many other works. Ourgoal is to determineand explain the propertiesof examples like (2a), where suru occurs with a direct object NP. 2 Shita is composedof the stem shi and a past markerta. We gloss all forms of the Verb as 'suru'. The entire direct object of suru will always be italicized, and the 0-markingNoun will be capitalized. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 207 The contributionof the object NP to 0-markingcan be easily seen in examples like (3a-c), where a differentarrayof argumentsappearsin each case. John-wa Mary-niHANASHI-o shita. suru John-TopMary-totalk-Acc 'John talked to Mary.' b. John-wa Tookyoo-karaSHUPPATSU-o shita. John-TopTokyo-from departure-Acc suru 'John departedfrom Tokyo.' c. John-wa murabito-ni[ookami-gakuru-to] KEIKOKU-oshita. John-Topvillager-to wolf-Nom come-Compwarn-Acc suru 'John warnedthe villagersthat the wolf was coming.' (3) a. The argumentarrayof the sentences in (3) varies with the Noun headingthe direct object (hanashi 'talk' vs. shuppatsu 'departure'vs. keikoku 'warning'). Suru has no influence on the numberand type of arguments.Setting aside some complicatedquestions regardingthe natureof argumentstructurein derived nominalsdiscussed in Grimshaw (1986; to appear),the same set of argumentsis seen to occur in nominalizations, where suru is not present and all the argumentsare NP-internal.(The postposition -ni may not occur before -no. In these and later examples we replace it by -e inside NPs.) John-no Mary-e-no HANASHI John-GenMary-to-Gentalk 'John's talk to Mary' b. John-no Tookyoo-kara-noSHUPPATSU John-GenTokyo-from-Gendeparture 'John's departurefrom Tokyo' c. John-no murabito-e-no [ookami-gakuru-to]-no KEIKOKU John-Genvillager-to-Gen wolf-Nom come-Comp-Genwarning 'John's warningto the villagersthat the wolf is coming' (4) a. The argumentstructureof the Noun licenses the argumentarray that occurs with suru, even thoughthe argumentsoccur outside the NP. The Noun is apparentlyassigning 0-roles outside its own maximalprojection.We will refer to such NPs as 0-transparent NPs. It is easy to show that in (3) all argumentsare outside the object NP. They have the case markingof the verbal/sententialsystem and not the nominalcase marker-no, contrastingwith the argumentsin (4). They can undergo scrambling:for example, the NP markedwith -ni in (3a,c) can be placed after the NP markedwith -o, giving a word order that is normalinside S but impossibleinside NP, where the head Noun is always the last element. We use the termlight Verbto referto Verbslike suruthat cooccur with 0-transparent NPs. Other Japanese Verbs are heavy and take only 0-opaque objects. For example, with wasureru 'forget' in (5), all argumentsof the Noun must appearinside the object This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 208 JANE GRIMSHAW AND ARMIN MESTER NP. Example (5a), where the argumentMary occurs inside the NP, is grammatical,but (5b), where Mary occurs outside and is markedwith -ni, is ungrammatical. (5) a. John-wa Mary-e-no HANASHI-o wasureta. John-TopMary-to-Gentalk-Acc forgot 'John forgot the talk to Mary.' b. *John-wa Mary-niHANASHI-o wasureta. John-TopMary-totalk-Acc forgot In sum, only suru takes a 0-transparentobject. Other verbs take only 0-opaque objects, which allow only inside arguments.Othergrammaticalpropertiescorrelatewith the difference between 0-opaque and 0-transparentNPs. Whereas 0-opaque NPs can freely be topicalized, 0-transparentNPs resist topicalization. This is illustratedin (6) and (7). (6) a. b. (7) a. b. 0-opaque object, topicalizationpossible John-ga [[ookami-gakuru-to]-no HOOKOKU]-owasureteita. John-Nom wolf-Nom come-Comp-Genreport-Acc had forgotten 'John had forgottenthe reportthat the wolf was coming.' [[Ookami-gakuru-to]-no HOOKOKU]-waJohn-ga wasureteita. wolf-Nom come-Comp-Genreport-Top John-Nomhad forgotten 0-transparentobject, topicalizationimpossible John-ga [ookami-gakuru-to] HOOKOKU-oshita. John-Nom wolf-Nom come-Compreport-Acc suru 'John reportedthat the wolf was coming.' *HOOKOKU-waJohn-ga [ookami-gakuru-to] shita. report-Top John-Nom wolf-Nom come-Compsuru The 0-opaqueobject of 'forget' in (6a) can be topicalized, as in (6b). The object NP in (7a) is 0-transparent,since an argumentof its head-the complementsentence ookamiga kuru-to 'that the wolf is coming'-is not markedwith -no and hence is not part of the NP. The ungrammaticalityof (7b), the topicalized version of (7a), shows that topicalization of 0-transparentNPs is not possible. There are other differences, which probablyreflect the nonreferential,predicatelike characterof transparentNPs. Only opaque NPs relativize, for example, and only opaque NPs allow modificationby elements like numerals. Nonetheless, the basic syntax of opaque and transparentNPs is identical. Both can containa sequence of prenominaladjectivesandothermodifiers,andboth reorderwithin S, following the standardJapanesepattern.Case markingis identical:both NPs receive accusative case, realized by -o. Both opaque and transparentNPs allow argumentsof the Noun to appearwithin the NP. In (8) one argument(murabito-ni)of keikoku'warn' appears outside the NP. The second argument(the complement sentence ookami-ga kuru-to)is inside the NP and is markedwith -no, just as it was in example (6a). This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 209 (8) John-wa murabito-ni[[ookami-gakuru-to]-no KEIKOKU]-oshita. John-Topvillager-to wolf-Nom come-Comp-Genwarn-Acc suru 'John warnedthe villagersthat the wolf was coming.' A complicatingfactor in this study is the existence of anothercase of suru, in which it takes opaqueobjects; in fact, it is a 0-markerof the usual kind. The ambiguitybetween this "heavy" suru and light suru is very similarto that between the English auxiliary do and the mainverb do. The object of heavy suru shows none of the characteristicsof transparentNPs; for example, topicalizationis possible with heavy suru, as shown in (9), where Tookyoo-e-noryokoo-wais a topicalized object. (9) [Tookyoo-e-noRYOKOOI-waJohn-ga shita. Tokyo-to-Gentrip-Top John-Nomsuru 'John made the trip to Tokyo.' The object of heavy suru does not allow outside arguments.This is why (7b) is not grammatical,with heavy suru as the verb. Heavy suru also places thematicrestrictions on the subject, which must be, roughly speaking, agentive. As a result, both topicalizability and resistance to outside argumentscorrelatewith agentivity. (lOa)is ungrammaticalbecause topicalizationrequiresheavy suru, but the subject (densha 'train')violates the agentivity requirement.(lOb) is grammaticalbecause it involves light suru (Oosaka-niis an outside argument)and light suru places no requirementson its subject. (10) a. *[Oosaka-e-no TOOCHAKU]-wadensha-ga shita. Osaka-to-Genarrival-Top train-Nomsuru 'The train arrivedin Osaka.' b. Densha-waOosaka-niTOOCHAKU-oshita. train-Top Osaka-to arrival-Acc suru Of course, both cases of suru are possible when the subject is agentive. Because of the differences between light and heavy suru, disambiguationcan usually be achieved with nonagentive subjects and/or outside arguments.Some contaminationof grammaticality judgments from the ambiguityis unavoidable,however. Speakers of Japanese are uncertainabout the status of a numberof the examplesbelow markedwith ?, judgingthem bad but finding it difficultto decide how ungrammaticalthey are. We interpretthe indecision as meaningthat the examples are ungrammatical,the judgmentsbeing blurred by the grammaticalityof the very same NPs as opaque objects. 2. 0-Markingwith Light Verbs In this section we will sketch our basic proposal for light suru. Our hypothesis is that suru is a Verb with only a skeletal argumentstructureand that 0-markingwith light suru is a result of suru combiningwith a 0-assigningNoun. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 210 JANE GRIMSHAW AND ARMIN MESTER 2.1. 0-Marking Nouns Since the transparentNP occurringwith suru is the source of 0-markingfor the clause, the Noun that heads such an NP must itself be a 0-role assigner. The Nouns hanashi 'talk', shuppatsu'departure,'andkeikoku'warning'resembletheirEnglishcounterparts, which can be argument-taking.Concrete Nouns like 'dog' and 'chair' do not have this capacity and thereforecannot head transparentNPs. This accords with the position of Anderson (1983-4, 5) that concrete and abstract Nouns differ in their 0-assigningcapacities, with only abstractNouns actingas 0-assigners.This idea is developed in Grimshaw (1986;to appear)and Safir(1987), where it is arguedthat "process" nominalsare 0-markersand "result" nominalsare not. Only process (that is, 0-marking)Nouns will cooccur with light suru. The differencebetween the two types is detectable in opaque nominals as well as transparentones. Fundamentally,the predictionis that Nouns that cannot head transparent NPs will not take argumentsin opaque nominals. They will occur only with possessives and other modifiers.The predictionappearsto be borne out; citing pairs of examples like those given in (3) and (4), Inoue (1976, 242-243) makes the important observation that the correspondenceis regular:every Noun that combines with light suru can also head a derivednominal.This generalizationcan be rephrasedin our terms: all Nouns that can head 0-transparentNPs can also head opaque process nominals, assigning the same 0-roles in each case. So the argumentstructureassigned to Nouns that head transparentNPs can be independentlymotivatedby opaqueprocess nominals. 2.2. Light Verbs Whatpropertyof suru makes it a light Verb? Ourhypothesis is that suru is thematically incomplete. In fact, it assigns no 0-roles; any 0-markingmust be done by anotheritem. A "light Verb," then, is one whose argumentstructureis skeletalor incomplete.In fact, it appears that suru assigns no 0-roles at all. Althoughit is a main Verb, its argument structureis more like that of an auxiliary:English do, for example. It is importantto stress here that light suru is not intrinsicallyagentive (see, for example, (lOb)), and it imposes no restrictions on the 0-role of its subject. No other selectional effects are detectable either: subjects of suru can be humanor nonhuman,animate or inanimate, and so forth, as illustratedin (11). John-ga Bill-to HANASHI-o shiteiru. suru John-NomBill-withtalk-Acc 'John is talkingto Bill.' mato-ni MEICHUU-o shita. b. Ya-ga suru arrow-Nomtarget-tohit-Acc 'The arrowhit the target.' (11) a. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS c. AND 0-MARKING 211 America-ga 200-nen-mae-niIgirisu-kara DOKURITSU-o shita. America-Nom200-year-ago-atEngland-fromindependence-Accsuru 'Americabecame independentof England200 years ago.' The Nouns with which suru occurs can themselves have any numberof arguments and any type of argumentstructure.The only positive propertyof suru is that it assigns accusative case; it is transitive. Of course, its direct object is not an argument-suru assigns case to a transparentNP, which bears the accusative case marker -o. Suru functions as a bearerof verbal inflectionfor the clause and as a case assigner, allowing the Noun in its direct object to assign 0-roles in a verbalcontext. Combiningan NP with suru turns the head Noun into the functionalequivalent of a Verb. The lexical entryof suruis given in (12). We use parenthesesto indicatethe argument list of the Verb: empty in the case of light suru. The notation (acc) indicates that suru assigns accusative case, but not to an argumentposition. (This is drawnfrom the Lexical FunctionalGrammartreatmentof raisingto object/exceptionalcase marking,where the verb is analyzedas takinga directobject thatdoes not correspondto any of its arguments (Bresnan (1982)).) (12) suru, V; ( ) (acc) As we have alreadysuggested, light suru resemblesin many ways the do of English Do Support, which carries inflection but assigns no 0-roles and imposes no selectional restrictions.The key differencebetween do and suru is that suru is transitive, so it can combine with NPs for 0-role assignment,whereas do must combine with anotherVerb. Note that the intransitivityof do supportsthe idea that the transitivityof suru is a lexical property, since there is lexical variationamong light Verbs. In sum, the head of the direct object NP is a 0-assignerbut does not carryinflection or assign accusative case. The Verb suruis not a 0-assigner,but it does assign accusative case and it does carry inflection. 3. The Distributionof Arguments 3.1. Transfer and 0-Marking The next issue, then, is how 0-markingworks in the NP-suru construction. As noted, when an NP is 0-transparent,the Noun's argumentstructureappears to be availablefor 0-markingargumentsoutside the NP. We will arguethatappearanceshere are misleading and that 0-markingin the surucomplexis local,just like 0-markingelsewhere. Arguments inside the NP are 0-markedby the Noun, and argumentsunderS are 0-markedby suru, which absorbs argumentstructurefrom the Noun. The basis of this is a process of Argument Transfer,throughwhich suru acquires a 0-markingcapacity. Thereare two logicallypossible cases: one where the Noun retains some of its own argument-taking capacity,andone whereit completelyloses this capacity and suru is responsiblefor all 0-marking. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 212 JANE GRIMSHAW AND ARMIN MESTER (13) illustratesthe first possibility of ArgumentTransferas it applies to suru and the Noun keikoku'warning',which takes an Agent, a Theme, and a Goal. Althoughwe will have reason to modify this later, for now we represent the argumentstructureof keikokuas a simple argumentlist. Transferremoves argumentpositions from the list associated with keikoku,insertingthem into the argumentstructureof suru and leaving only the Theme in the argumentstructureof keikoku. (13) a. keikoku(Agent, Goal, Theme) b. suru ( ) (acc) c. keikoku(Theme) + suru (Agent, Goal) (acc) (13a,b) are the inputs to ArgumentTransfer, and (13c) is the result, a pair of lexical items (which must appear together). The Noun keikokuretains just the Theme role, which will be assignedNP-internally,andthe transitiveVerb suruassigns the transferred roles Agent and Goal outside NP. These argumentstructuressupport the 0-markingschematicallyindicated in (14), where the 0-role of each phrase is indicatedin squarebrackets. (To simplifythe representation, we do not include a VP node in our diagrams.)This is how 0-markingworks for (15) (= (8)). (14) [Agent] [Goal] S' [Theme] keikoku (theme) shita (Agent, Goal) (acc) (15) John-wa murabito-ni[[ookami-gakuru-to]-no KEIKOKU]-oshita. John-Topvillager-to wolf-Nom come-Comp-Genwarn-Acc suru 'John warnedthe villagersthat the wolf was coming.' The second possibility is for suru to absorb all argumentsof the Noun, leaving the Noun with no 0-markingcapacities. In this case all argumentsare 0-markedby the Verb and are thereforerealized outside the object NP. This is illustratedin (16)-(18). (16) a. keikoku(Agent, Goal, Theme) b. suru ( ) (acc) c. keikoku( ) + suru (Agent, Goal, Theme) (acc) The derived argumentstructureof suru 0-marksall the arguments, which occur outside the transparentNP as in (18). This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 213 0-MARKING S (17) NP-wa [Agent] NP-ni S' [Goal] [Theme] NP-o I keikoku V shita (Agent, Goal, Theme) (acc) KEIKOKU-oshita. (18) John-wa murabito-ni[ookami-gakuru-to] John-Topvillager-to wolf-Nom come-Compwarn-Acc suru 'John warnedthe villagersthat the wolf was coming.' The result of Transferis a pairof linkedlexical items that must be insertedtogether. Obviously, no well-formedstructurewould result if an argumentstructurefrom (13c) was combinedwith an argumentstructurefrom (16c). Insertingsuru with the argument structurein (13c) togetherwith keikokuwith the argumentstructurein (16c) would result in no Theme being realized;insertingsuru with the argumentstructurein (16c) together with keikokuwith the argumentstructurein (13c) would result in two Themes being realized. The lexical insertionof the Noun-Verbcomplex can presumablybe assimilated to a more general theory of phrasal constructions, which will govern the behavior of idioms (like kick the bucket) and other lexical expressions that do not constitute single words. In support of the hypothesis that the Noun and suru are listed together in a derived lexical entry we can cite the fact that there is lexical variationamong Nouns with respect to the suru construction. For example, shooshin 'promotion'occurs with suru in the incorporationconstruction, as in (19a), and can head an opaque NP, as in (19b). However, shooshin in a phrasalsuru complex (illustratedin (19c)) is accepted by some speakers and rejected by others. (19) a. John-wa buchoo-ni SHOOSHIN shita. John-Topsection chief-to promotion suru 'John obtainedpromotionto section chief.' b. Buchoo-e-no shooshin-ga hayakatta. section chief-to-Genpromotion-Nomfast-Past '(He) won speedy promotionto section chief.' c. ??John-wa buchoo-ni SHOOSHIN-o shita. John-Topsection chief-to promotion-Accsuru 'John obtainedpromotionto section chief.' There appears to be no principledreason why shooshin should be ill-formedin (19c), so perhapsthis representsa lexical gap for some speakers. An immediatequestionis why Transfercannot occur withjust any Verb-Nounpair, instead being limited to light Verbs like suru. Our suggestion is that Transferrequires an incomplete argumentstructure-one with "space" for the addition of arguments. Verbs with completely specified argumentstructures cannot be targets for Transfer. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 214 JANE GRIMSHAW AND ARMIN MESTER Transferdoes not requirea completelyempty argumentstructure-suru has a causative form, saseru, which is illustratedin (20). saseta. (20) Mary-ga John-ni Bill-to AISEKI-o Bill-with Mary-NomJohn-to table-sharing-Accsuru-cause 'Marymade John share a table with Bill.' Ouranalysisof (20)is this: causativizationadds an argumentto suru, givingit the partially specified representationin (21). (21) saseru (Agent, ) (acc) Transferthen adds argumentsin the usual way. This is possible because the argument structureof saseru is incomplete, so it remainsa legitimatetargetfor Transfer. A crucial feature of the theory proposed here is that there are two 0-markersin a suru complex, and two separatedomainsof 0-marking.Transferdoes not form a single predicate; it forms two predicates, each with altered 0-markingproperties. We could instead try to analyze these constructions as involving only a single 0-marker,either suru or the Noun actually being responsible for all 0-marking.If, for example, suru always inheritedall argumentsfrom the Noun, it could 0-markboth inside and outside arguments,and the Noun would not be a 0-markerat all. Alternatively,suru could be viewed as somehow allowingthe Noun to 0-markthroughthe NP node, so that the Noun could 0-markat the S level as well as within NP. The Noun would then be behaving essentially as though the NP node were absent. In both of these analyses there would be only one 0-markerand one 0-markingdomain. The argument-transfer theory, with its two 0-markers,has two importantadvantages over a theory that posits a single 0-marker.We summarizethe relevantpoints here and provide a more detailed discussion in section 4. (i) This solution maintainsstrict locality of 0-marking,since 0-markingin the two domains, NP and S, is performedby two differentargumentstructures.Single-predicate theories would necessarily involve nonlocal 0-marking(a case of (lb) or (lc), in fact). (ii) In sections 3.2-3.4 we will explore the behavior of argumentswithin the suru complex. Part of the explanationfor their distributiondepends crucially upon the idea that an argumentinside NP is 0-markedby the Noun alone, whereas an outside argument is 0-markedby the suru predicate. To sum up, in this theory NP and S still constitute distinct domainsfor 0-marking even for transparentNPs. Argumentsof the Noun can go inside or outside the direct object NP because they can receive their 0-markingequally well in either position, satisfying the argumentstructureof either the Noun or the suru predicate. 3.2. Transfer and Outside Arguments As articulatedso far, the theory of light Verbs predicts that argumentsof the Noun can freely occur distributedinside or outside the direct object NP. In fact, the distribution This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS 215 AND 0-MARKING of argumentsis not completely free. There are three importantgeneralizations: (i) At least one argumentapartfrom the subject must be outside the NP. (ii) The subject argumentmust always be outside the NP. (iii) For Nouns that take a Theme and a Goal, if the Theme argumentis realized outside NP, the Goal must also be realized outside NP. We will argue that these restrictionsfollow from the hierarchicalorganizationof argument structureplus the assumptionthatTransfermusttransmitat least one unsuppressed argumentposition to the suru predicate. For Nouns that take three arguments(a subject and two others) both nonsubject argumentscan be realized outside the NP, as in (22a). Alternatively,one can be inside and one outside, as in (22b), a phenomenonwe will refer to as splitting. However, it is not possible for both to appearinside the NP, even when the comparableopaque NP is fully well-formed.This is schematizedin (22c). S (22) a. b. NP NP NP NP-o V suru c. * S NP NP NP-o NP N V suru S NP NP-o NP NP N V suru For example, Nouns like shoomei 'proof combine with suru and take both a sentential complementand an indirectobject. (23a) shows that both argumentscan occur outside, leavingjust shoomei-o as the direct object NP. (23) a. Sono deeta-ga wareware-ni[kare-noriron-ga machigatteiru-to] that data-Nomus-to he-Gen theory-Nom mistaken be-Comp SHOOMEI-oshiteiru. proof-Acc suru 'That data proves to us that his theory is mistaken.' b. Sono deeta-ga wareware-ni[[kare-noriron-ga machigatte that data-Nomus-to he-Gen theory-Nom mistaken SHOOMEI]-oshiteiru. iru-to]-no suru be-Comp-Genproof-Acc This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 216 JANE GRIMSHAW AND ARMIN MESTER machigatte c. ?Sono deeta-ga [wareware-e-no[kare-noriron-ga he-Gen theory-Nom mistaken that data-Nom us-to-Gen SHOOMEI]-oshiteiru. iru-to]-no suru be-Comp-Genproof-Acc In (23b) the sententialcomplementoccurs inside the object NP, as indicatedby the -no marker.The indirect object wareware-ni'to us' is outside, however, so here the arguments of the Noun are split between the NP and the clause. (23c) is ungrammatical,with both the indirectobject and the sententialcomplementinside NP. It mightbe suggested that the decline in grammaticalityof (23c) should be attributedto the substitutionof -e for -ni mentionedearlier.However, Naoki Fukuihas pointed out to us that the contrast between (23a,b) and (23c) is preserved even if ni is replaced by nitaishite, which can occur inside NPs as well as in clauses. The prohibitionagainst having both argumentsinside NP cannot be reduced to a general condition governing NPs. The reason is that (24), the opaque counterpartto (23c), is well-formed,if complex. machigatte (24) [Sono deeta-no wareware-e-no[kare-noriron-ga he-Gen theory-Nom mistaken that data-Genus-to-Gen iru-to]-no SHOOMEI]-otsukatta. used be-Comp-Genproof-Acc '(I) used that data's proof to us that his theory is mistaken.' Other examples support the general conclusion. Shisa-o suru 'to suggest' takes a direct object and an indirectobject. For reasons irrelevantin the present context, both argumentscannot be outside with normalcase marking:(25) is ungrammaticalbecause it violates the Double -o Constraint(Harada(1973), Shibatani(1973), Kuroda (1978), Saito (1985)). (25) *Sono hookokusho-waMary-ni[kaiketsu-no hookoo]-o that report-Top Mary-to solution-Gendirection-Acc SHISA-o shiteiru. suggestion-Accsuru 'That report suggests to Mary the directionof the solution.' Topicalizingkaiketsu-nohookoo 'directionof the solution' circumventsthe Double -o Constraint,and (26a) shows that it is indeed possible to have both argumentsoutside. As before, it is perfectlyacceptableto have one argumentinside andone outside, splitting the two argumentsbetween the NP and the clause, as in (26b). To have both arguments inside is not possible, however, so (26c) is ungrammatical. (26) a. [Kaiketsu-no hookoo]-wa sono hookokusho-gaMary-ni solution-Gendirection-Topthat report-Nom Mary-to shiteiru. SHISA-o suggestion-Accsuru This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 217 b. Sono hookokusho-waMary-ni[[kaiketsu-no hookoo]-no that report-Top Mary-to solution-Gendirection-Gen SHISA]-o shiteiru. suggestion-Accsuru c. ?Sono hookokusho-waMary-e-no [kaiketsu-no hookoo]-no that report-Top Mary-to-Gen solution-Gendirection-Gen SHISA]-o shiteiru. suggestion-Accsuru The opaque NP correspondingto (26c) is grammatical. (27) Bill-wa [sono hookokusho-noMary-e-no [kaiketsu-no hookoo]-no Bill-Top that report-Gen Mary-to-Gen solution-Gendirection-Gen SHISA]-o mushishita. suggestion-Accignored 'Bill ignoredthat report's suggestionto Maryof the directionof the solution.' In general, then, light Verb complexes with two nonsubjectargumentsallow both argumentsto be outside the object NP, and they allow splittingbetween the inside and outside positions, but they do not allow both argumentsto be inside. This restriction appears to be specific to transparentNPs, given that opaque NPs allow multipleinside arguments. One way of stating the descriptive generalizationis to say that one nonsubject argument must be outside the NP. This connects the behavior of Nouns with three argumentsto an apparentlydifferentrestrictionfound in complexes with two arguments. In such cases no argumentat all can appearinside the NP, and therefore only outside argumentsare found. This is exemplifiedby the locative argumentof toochaku 'arrival' in (28), which can appearoutside the NP, but not inside. (28) a. Densha-waUenoeki-ni TOOCHAKU-oshita. train-Top Ueno station-toarrival-Acc suru 'The train arrivedat Ueno station.' b. ?Densha-wa[Uenoeki-e-no TOOCHAKU]-o shita. -to-Gen -Acc Similarly,the complementsof shuppatsu 'departure'and aiseki 'table-sharing'must be outside. All three take inside argumentswhen they head opaque NPs, as illustratedfor aiseki 'table-sharing'in (31). (29) a. Densha-wa Uenoeki-kara SHUPPATSU-o shita. train-Top Ueno station-fromdeparture-Acc suru 'The train departedfrom Ueno station.' b. ?Densha-wa[Uenoeki-kara-noSHUPPATSU]-o shita. (30) a. John-wa Bill-to AISEKI-o shita. John-TopBill-withtable-sharing-Accsuru 'John shareda table with Bill.' This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 218 JANE GRIMSHAW AND ARMIN MESTER b. ?John-wa [Bill-to-no AISEKI]-o shita. John-Top Bill-with-Gentable-sharing-Accsuru (31) John-wa [Bill-to-no AISEKI]-o kotowatta. John-Top Bill-with-Gentable-sharing-Accrefused 'John refused to share a table with Bill.' Ourconclusionis thatat least one argument(otherthan the subject)must be outside the NP. As a result, three-argumentcomplexes allow at most one inside argument, whereas two-argumentcomplexes allow no inside arguments.Setting aside for the moment the case of the subject, this means that at least one position in the argument structureof the Noun must be transferredto suru. This generalizationhas a numberof furtherconsequences. First, a Noun with three nonsubjectargumentsshouldallow two of themto occur inside, leavingonly one outside. (32) is a candidateexample, and the judgments supportthe predictions. (32a) contains only one inside argument,(32b)contains two inside arguments,and (32c) contains three inside arguments. (32) a. America-wa kin-de Mexico-to [shinamono-noTORIHIKI]-o America-Topgold-forMexico-with goods-Gen business-Acc shiteiru. suru 'Americadoes business in goods with Mexico in exchange for gold.' b. America-wakin-de [Mexico-to-noshinamono-noTORIHIKI]-oshiteiru. c. ?America-wa[kin-de-noMexico-to-noshinamono-noTORIHIKI]-o shiteiru. As expected, it is ungrammaticalto include all three nonsubjectsin the NP, as in (33a), although the opaque nominalin (33b), which correspondsto the ungrammaticaltransparent NP, is well-formed. (33) a. ?America-wa[kin-de-noMexico-to-noshinamono-noTORIHIKI]-o shiteiru. b. Kin-de-no Mexico-to-no shinamono-noTORIHIKI-waabunai. gold-for-GenMexico-with-Gengoods-Gen business-Top dangerous 'Doing business with Mexico in exchange for gold is dangerous.' A second prediction is that only arguments should be relevant for determining whether an argumenthas been transferredto suru. Adjuncts, since they are not listed in the argumentstructure,cannot be transferred.An adjunctappearingoutside the NP should thereforehave no effect on grammaticality.This predictionseems correct: (34a) and (34b)correspondto the ungrammatical(23c) and (26c). Both are ungrammaticaleven though an adjuncthas been added in the outside position (althoughspeakers do report a slight improvement,inexplicablein our terms). This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 219 0-MARKING (34) a. ?Sono deeta-ga koko-de/kyoo [wareware-e-no[kare-no that data-Nomhere/today us-to-Gen he-Gen riron-ga machigatte iru-to]-no SHOOMEI]-o shiteiru. theory-Nommistaken be-Comp-Genproof-Acc suru b. ?Sono hookokusho-wasaigoni/koko-de[Mary-e-no that report-Top finally/here Mary-to-Gen [kaiketsu-no hookoo]-no SHISA]-o shiteiru. solution-Gendirection-Gensuggestion-Accsuru So at least one argumentof the Noun must transferto suru. Why? We suggest that when no Transferoccurs, the 0-Criterionis violated. DuringTransfer,which is after all a kind of complex predicateformation,the NP becomes exempt from the 0-Criterion. It is certainlynot an argumentof suru, so it cannotbe 0-marked.Presumablyit is licensed in a differentfashion-by participationin 0-assignment.If no Transferoccurs, however, the NP has the usual status and must be construedas the argumentof some predicate. Suru is light and cannot meet this requirement;hence, a violation ensues. In this way, we can derive the result that one argumentmust be realized outside the object NP, as a side effect of Transfer. 3.3. Subjects There are two issues to be addressedconcerningthe behaviorof subjects in suru complexes. The first is why the subjectof the complex must always be outside the NP. The second concerns the properties of Transfer:since the subject appears outside NP, it must be 0-markedby suru. In orderto be 0-markedby suru, the subjectmust have been transferredto the suru predicate. Yet we have just shown that an additionalargument must also be transferred.The problemis, then, why transferringjust the subject does not exempt the NP from the 0-Criterion. There are many well-knowndifferencesbetween subjects of Nouns and subjects of Verbs; it has often been assumedthat Nouns and Verbs have the same kind of argument structurebut are subject to differentprinciplesgoverningthe realizationof their arguments. In this category are the proposals of Anderson (1983-4), Kayne (1981), and Rappaport(1983).However, recent work on nominalizationsuggests that the "external" argumentstructurepositioncorrespondingto the subjectof a Verbis lexically suppressed for Nouns (see Zubizarreta(1985),Grimshaw(1986;to appear)),ratherthan internalized as suggested in Williams(1981). As a result, the argumentposition can never be syntactically satisfied: its status is very differentfrom that of other argumentsof Nouns and Verbs. Possessive NPs, then, are adjuncts,ratherthanarguments,a view developed more fully in Grimshaw(1986;to appear).A related hypothesis, based on the work of Esther Torrego on the ECP and extractionfrom NP, can be found in Chomsky (1986, 45-46). For the sake of concreteness, we can view 0-markingas assignment of an index from a position in an argumentstructureto the correspondingphrase, roughly as in This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 220 JANE GRIMSHAW AND ARMIN MESTER Stowell (1981).We can define an open positionas one whose index has not been assigned to any phrase. A suppressedargumentis an argumentposition with no index to assign. For example, shoomei 'proof takes three arguments:Agent/Source, Theme, and Goal. Since it is a Noun, the Agent/Source(subject)position is suppressedas in (35) and has no associated index. (35) shoomei (Agent/Source,Themej, Goalk) We have established that Transferstrips positions from the argumentstructureof the Noun and assigns them to that of suru. If the subject of the Noun were an open position,just like any otherargumentposition,then it shouldcertainlycountfor Transfer, and no other argumentshould have to be transferred.However, if the real requirement is that an open position must be transferred,the suppressedargumentwill not count for the computation,andthe desiredresultwill follow. It seems, then, thatthe basic principle governing0-transparentNPs is that the Noun must transferto suruat least one argument with an unassigned index.3 One possible interpretationof this is that more generally, suppressed arguments may not be visible for argument structure operations, which shouldperhapsbe viewed as applyingto a representationin which suppressedarguments are omitted entirely. We returnnow to the first restrictionto be explained:the fact that the subject must always appear in the clause and not within the NP, that is, as an outside and not an inside argument.For example, (36b) is ungrammatical,even though the Goal (Mary-ni) is 0-markedby the suru predicate, so that Transferhas apparentlyoccurred. Moreover, the NP itself in (36b)is not structurallyill-formed:the opaquenominalin (36c)is identical, and grammatical. (36) a. John-ga Mary-niHANASHI-o shita. John-NomMary-totalk-Acc suru 'John talked to Mary.' b. *Mary-niJohn-no HANASHI-o shita. 3 If the Noun must always transferan open position to the Verb, and if the "subject" never counts as open for this purpose, no single-argumentNouns should ever participatein the suru construction.A oneargumentNoun will have only a "subject" argumentposition, and we know that this is not sufficientfor transfer.The predictionis difflcultto evaluate,becauseas we showedin section 2, suruis ambiguousbetween the lightVerb of interesthere, and a heavy Verb, ratherlike mainVerb do in English.The situationis further complicatedby the fact thatin the case of monadicNouns the techniquefor disambiguationthatwe have used throughout(outsidearguments)is unavailable.However, since heavy suru has agentive(or at least "actor") subjects, Nouns like those in (i) can presumablybe analyzedas complementsto heavy suru. (i) suiei-o suru'swim', sampo-osuru'walk', seppuku-osuru'commitsuicide', undoo-osuru'exercise', seikatsu-osuru 'makea living', shigoto-osuru 'work',jisatsu-o suru 'commitsuicide' Once agentiveNouns like these are factoredout, a few cases remain,includingthose in (ii). (ii) wakajini-osuru 'die young', byooki-o suru 'be sick', fuhai-o suru 'become rotten', daraku-osuru 'be morallyruined',kushami-osuru 'sneeze', seki-o suru 'cough', bimboo-osuru 'become poor' The analysis of these remainsuncertain.They may representa smallclass of fixed expressions,or they may be unaccusativeNouns, like the Verbs with similarmeanings(Perlmutter(1978),Burzio (1981)).In this case their single argumentwouldbe internalratherthanexternaland mightbe unsuppressedand thus transferable. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT c. VERBS AND 0-MARKING 221 John-no hanashi-o hometa. John-Gentalk-Acc praised '(I) praised John's talking.' Since an outside argumentis 0-markedby suruand an inside argumentby the Noun, the issue of why the subjectmust be outside reduces to the question of why the subject argumentmust be transferredto suru and cannot remainas an argumentof the Noun. If it could remainas an argumentof the nominal, it would remainlexically suppressed. Linkinga genitive phraseto it wouldpose no problem,and (36b) shouldbe grammatical. If, on the other hand, the argumentis transferredto suru, it will be the argumentof a Verb, not a Noun. We assume that it will therefore be activated and will receive an index. Since it is 0-markedby the suru predicate,it must be under S, not inside the NP. (The fact that the argumentthat is suppressedfor the Noun is reactivatedfor the Verb provides crucial evidence that the suppressed argumentmust be represented in the Noun's argumentstructurein some form. Otherwise, Transferbased on the argument structureof the Noun could not possibly have the right effect.) Why should the subject argumentalways transfer?So far we have establishedthat one open argumentmust be transferredto suru. Of necessity, this will always be an internal argument,in the sense of Williams (1981), since only internal argumentsare unsuppressed for Nouns. The desired result will follow if transferringthe internal argumenthas transfer of the external argumentas a consequence. Since the internal argumentmust always be transferredto avoid a violation of the 0-Criterion,it will then follow that the external argumentmust be too. We propose that this dependencybetween the external and internalargumentsfollows from the organizationof argumentstructure,in accordancewith the usual division into external and internal arguments,based on Williams (1981). Like Hale (1983), we take the external/internaldistinction to be reflected in the organizationof argument structure, with the external argument-the Agent/Source of shoomei, for example being external, and representedas more prominentthan the other arguments.The argument structureof shoomei is given in (37). (37) shoomei (Agent/Source(Goalj, Themek)) Transferacts in a top-downfashion,preservingthe structuralorganizationor prominence relationsof the argumentstructure:it cannotremove a lower argumentwithoutremoving all the higher argumentsas well. This will disallow an outcome in which the Goal has been transferredand the Agent has not, as in (38a), but will allow one in which both have been transferred,as in (38b). (38) a. b. suru (Goalj) + shoomei (Agent, Themek) suru ((Agenti, Goalj)) + shoomei (Themek) Crucially, then, the process of ArgumentTransfermust apply in an outside-infashion, thus preservingthe basic organizationof the input argumentstructure.Given this, plus This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 222 JANE GRIMSHAW AND ARMIN MESTER the hypothesis that the Agent/Source is the "highest" argument,the situation can be fully explained. There are, inevitably, some alternativeaccounts for the positioningof the subject in the suru complex. Let us examinethe situationthat would arise if suru appearedwith a direct object NP and no transferhad occurred. Schematically,we would have a configurationlike (39), instantiatedin (40). (39) 5 NP-o NP ... NP (40) *John-no Bill-to-no V N suru AISEKI-o shita. John-GenBill-with-Gentable-sharing-Accsuru One line of reasoningmight appeal to the argumentstatus of the subject. Perhaps the subject is really an argumentof suru and not of the Noun, and this is why it cannot appearinside the NP. There are two problemsfor such an account. First, suru imposes absolutely no requirementson this NP, so it is not an argumentin the usual sense. To call it one is tantamountto subcategorizingfor a subject. Second, this proposal would allow for other light verbs differingfrom suru in precisely this respect: they would not have such an argumentslot and would thereforeallow the subjectof the Noun to appear withinNP. Presumably,though,it is not an accidentallexical propertyof suruthat makes inside subjects impossible. Another possibility is that it is the absence of a subject for the clause that makes the configurationin (39) impossible, if (39) violates some condition (like the Extended Projection Principle of Chomsky (1982)) that requires clauses to have subjects. The natureof the explanationhere is not clear-presumably the clause could contain a null expletive subject, satisfying the structuralcondition and leaving the real argumentof the Noun free to occur inside. Moreover, there is another problemfor an explanation based on the absence of a subject. Recall that saseru, the causative of suru discussed in section 3. 1, has one argumentthat acts as the subjectof the entirecausativepredicate. We have alreadygiven an example with saseru in (20) (repeatedhere as (41)). (41) Mary-ga John-niBill-to AISEKI-o saseta. Mary-NomJohn-toBill-withtable-sharing-Accsuru-cause 'Marymade John share a table with Bill.' In (41) all argumentsof aiseki have been transferredto saseru and are realized outside the NP. Since saseru has a subject argument(Mary-ga)that is realized as the subject This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 223 of the clause, no Extended Projection Principle violation would result if the subject argumentof aiseki (John) was not transferredto saseru but instead was realized inside the NP. Nevertheless, failure to transfer the subject argumentof aiseki leads to ungrammaticality,regardlessof whether anotherargument(Bill-to) is also transferred,to comply with the obligatorinessof Transfer. Thus, both (42a) and (42b) are ungrammatical.4 (42) a. *Mary-ga [John-no Bill-to-no AISEKI]-o saseta. Mary-Nom John-GenBill-with-Gentable-sharing-Accsuru-cause b. *Mary-gaBill-to [John-noAISEKI]-o saseta. As usual, the corresponding0-opaqueNP is grammatical,showing that there is nothing intrinsicallywrong with the object NP in (42). (43) [John-no Bill-to-no AISEKI-wa] machigaidatta. John-GenBill-with-Gentable-sharing-Topmistake was 'John's sharinga table with Bill was a mistake.' In (42) saseru has a subject, but John must still be 0-markedby saseru and thus must be transferredfrom aiseki. Hence, it must appearoutside the NP, even though it is not acting as the subject of the suru complex. We conclude that the obligatory outside positioning of this argumentcannot be reduced to a condition requiringsubjects for clauses. We have arguedthat the behaviorof subjects in the suru complex follows from the representationof argumentstructure. We will show in the following section that this explanationgeneralizes to the behaviorof internalarguments. 3.4. The Distributionof InternalArguments The argumentsof a Noun can be split between the inside and outside positions. This was shown earlierby (23b) and (26b), repeated in (47a) and (48a). However, switching the positions of the argumentsmakes these examples ungrammatical.The data are presented schematicallyin (45) and (46). The Goal can be outside the NP with the Theme inside, but the Theme cannot be outside with the Goal inside. The relevant examples follow in (47) and (48). 4One complicationis that (42a) improvesconsiderablyif the subjectof aiseki is transferredwhile Billto stays behind. (i) ?Mary-ga John-ni[Bill-to-no AISEKI]-o saseta. Mary-NomJohn-to Bill-with-Gentable-sharing-Accsuru-cause No open position has been transferredin this example, which should thereforebe ungrammaticalas a suru complex. However, note that none of the criticalpropertiesof light suru are involved here, so this example may actually involve heavy suru. On the other hand, the marginalityof (30b) suggests that aiseki may not combine with heavy suru. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 224 AND ARMIN MESTER JANE GRIMSHAW S (45) NP [Source] NP-o NP [Goal] V I shita N S' [Theme] shoomei (46) * s NP [Source] S' [Theme] NP-o NP [Goal] (47) a. N V shita shoomei Sono deeta-ga wareware-ni[[kare-noriron-ga he-Gen theory-Nom that data-Nomus-to SHOOMEI]-o shiteiru. machigatte iru-to]-no suru mistakenbe-Comp-Genproof-Acc 'That data proves to us that his theory is mistaken.' b. *Sono deeta-ga [kare-noriron-ga machigatteiru-to] that data-Nom he-Gen theory-Nom mistaken be-Comp [wareware-e-no SHOOMEI]-o shiteiru. us-to-Gen (48) a. proof-Acc suru Sono hookokusho-wa Mary-ni [[kaiketsu-no hookoo]-no that report-Top SHISA]-o Mary-to solution-Gendirection-Gen shiteiru. suggestion-Accsuru 'That report suggests to Mary the directionof the solution.' b. *[Kaiketsu-no hookoo-wa] sono hookokusho-ga solution-Gendirection-Topthat report-Nom Mary-e-no SHISA-o shiteiru. Mary-to-Gensuggestion-Accsuru Note that both phrases at issue would normallybe taken to be argumentsof the Noun, and not merely adjuncts.As evidence for their argumentstatus we can cite the fact that the transferof the ni phraseis evidentlysufficientto makethe surucomplexwell-formed. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 225 Theme argumentsare not absolutely barredfrom the outside position: they occur outside in grammaticalsentences providedthat the other argumentis also outside ((23a) and (26a)). It seems highlyunlikelythat indirectobject argumentsare absolutelybarred from occurringinside the NP. After all, they certainly can occur inside opaque NPs. Moreover, in (32b), one of the four-argumentNoun examples, a rather similartype of argument-Mexico-to-no 'with Me-xico'-does occur inside, suggesting that Goals should be able to as well, under the right circumstances. It does not seem to be the absolute positions of the individualargumentsthat are responsiblefor the ungrammaticality of (47b) and (48b). Instead, the relative position is the importantmatter: the combinationof an outside Theme and an inside Goal is ill-formed. Why shouldthis combinationbe ill-formed?It appearsthat "closeness" to the Noun in some sense is the key factor here. When splitting occurs, the more closely linked Theme argumentsmust be inside, the less closely linked argumentsmust be outside. Our proposal is that the internalargumentsin an argumentstructureare hierarchically structured.Just as the external argumentis more prominentthan the internalarguments, so some internal argumentsare more prominentthan others. In particular, Goals are more prominentthan Themes, so the argumentstructureof shoomei 'proof is (49), where the parenthesesindicate the structureassigned to the argumentlist. The most prominentargumentis the Agent/Source argument,then the Goal, and then the Theme. (49) shoomei (Agent/Source(Goalj(Themek))) Withthis hypothesisaboutargumentstructurerepresentationthe theory of Transfer already given will explain the generalization.The restrictionon splittingfollows from the fact that the transferof argumentsrespects the structureof the inputrepresentation: the process can peel off outside layers but can never remove an inside layer without also removingthe layers outside it. This is exactly the principlethat was motivatedby consideringthe distributionof subjects in suru complexes. As our example, we will use an argumentstructurecontainingan Agent/Source, a Goal, and a Theme. The basic form is given in (50). (50) (Agent/Source(Goalj(Themek))) Transferyields a numberof possible outcomes when this argumentstructurecombines with suru. Recall that suru cannotjust receive the Agent/Sourceargument,because it is not an open position. Suru can receive the Agent/Source and the Goal, however, giving the result in (51), where suru 0-marksan Agent/Sourceand a Goal, and the Noun 0-marksa Theme. As discussed above, after Transferthe Agent/Sourceposition in the argumentstructurecarries an index since it is now part of a verbal argumentstructure and therefore is not suppressed. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 226 JANE GRIMSHAW (51) AND ARMIN MESTER 5 NP-ga [Source] NP-ni V NP-o [Goal] shita ,. N S' (Source (Goal)) [Theme] shoomei (Theme) The Transferinvolved here is legitimate,because it preserves the structureof the input. (52) shoomei (Agent/Source(Goal (Theme))) shoomei (Theme) + suru (Agent/Source(Goal)) For the ungrammaticalexamples (47b) and (48b), on the other hand, Transfercannot derive appropriateargumentstructuressince it would have to remove an argumentlower in the structure,withoutalso removingthe higherarguments.For (47b)the Themewould have to be transferred,illegitimatelyleaving the more prominentGoal behind, as shown in (53). (53) * NP-ga [Source] S S' [Theme] NP-o NP-e-no [Goal] V shita (Source (Theme)) N shoomei(Goal) 4 shoomei (Agent/Source(Goal (Theme))) shoomei (Goal) + suru (Agent/Source(Theme)) In fact, then, the ungrammaticalityof (47b)is due to a violation of the 0-Criterion:there is no suru complex that can 0-markboth the outside Theme and the inside Goal. In the grammaticalcases of splitting, the Theme is 0-markedby the Noun, and the Goal by suru. In this way, the behavior of internalargumentsin the suru complex follows from the theory of argumentstructurerepresentation.Provided that Transferrespects the structureof argumentstructure,the range of possible positions for Themes and Goals can be derived. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 227 4. An Alternative In the proposal made here, the suru complex involves two 0-markers,not just one. A key feature of the process involved is that, except when all argumentsare transferred, the result is not a single 0-marker.Neither suru nor the Noun is alone responsible for all 0-marking.Each has its own domain,as usual, but carriesan argumentstructurethat is modified by its role in the suru complex. It is illuminatingto compare this solution to an alternative in which there is only one 0-marker,and suru is responsible for 0markingall arguments,whether they are inside the NP or outside it. In this solution complex predicateformationwould transferall argumentsfrom the Noun to suru, leaving the Noun with no argumentsat all. An obvious disadvantageto this account is that it would requirenonlocal 0-markinginto the NP. Assumingthat this is allowed, how would the propertiesof the suru complex be explained?There are three problems of interest: the outside positioningof the subject, the relative positioning of internalarguments,and the requirementthat one argumentapartfrom the subject must occur outside. The explanationfor the outside positioningof the subject within this theory would depend on the idea that a possessive-markedNP is incapableof satisfyingan argument position of the complex predicate. Since Nouns do not have subject arguments, the -no-markedphrasesthat seem to satisfy those argumentpositions must reallybe adjuncts and not arguments;see Grimshaw(1986;to appear). However, the complex predicate that is formed by combiningsuru and a Noun is verbal in character,since suru itself is a Verb. Verbs do not have satisfied subjects;hence, the argumentstructureformed by complex predicateformationwill have an unsuppressedsubject argument,which must be syntactically satisfied by a phrase in the clause. Now we can see what would explain the ungrammaticalityof (54a), for example: John-no can never satisfy the Agent argumentof benkyoo-suru;it is not even an argument. Hence, the ungrammaticalityof inside subjects reduces to a violation of the 0Criterion. In (54b) the grammaticalcounterpartJohn-ga is 0-markedby the complex predicate. It is an argumentratherthan an adjunct;hence, no violation results. (54) a. *Nihongo-wa [John-no BENKYOO]-o shiteiru. Japanese-Top John-Genstudy-Acc suru b. Nihongo-wa John-ga BENKYOO-oshiteiru. Japanese-TopJohn-Nomstudy-Acc suru 'John is studyingJapanese.' The relative positioningof internalargumentsmight be derivablefrom the theory of 0-marking,given two assumptions.First, the argumentstructurerepresentationmust be as we are suggesting, with the Theme argumentlower than the Goal. Second, 0marking(ratherthan Transferas in the proposed theory) must respect the structureof the argumentstructurein the followingsense. Argumentslower in the argumentstructure must be 0-markedbefore argumentshigherin the argumentstructure.Providedthat 0- This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 228 JANE GRIMSHAW AND ARMIN MESTER markingproceeds in a bottom-upfashion, 0-markinginside NP will precede 0-marking outside NP. As a result, the argumentthat is lower in argumentstructure(the Theme) cannot be outside the NP, if the Goal is inside. Again, then, the restrictionmightreduce to a violation of the 0-Criterion.The exact characterof 0-markingin this proposal is a little suspect: it assumes that the NP could define a domain of 0-marking,despite the fact that it is transparentand the 0-markeritself is outside it. The proposalalso assumes that the Verb can assign a 0-role within the NP domain, despite the fact that the Verb itself is outside that domain.The notion of "domain"for 0-markingis thus disconnected from the relationshipof an argumentto a 0-marker,since argumentsof a single predicate appear in differentdomains. Whetherthese points renderthe solution untenableis unclear to us. Setting this question aside, so far the single 0-markertheory seems to be able to matchour proposal.A problemarises, however, with the requirementthat one argument must go outside, in additionto the subject. It is hardto see how this mightbe expressed, let alone explained,if the surucomplex containsa single 0-marker.Once we have stated that one nonsubjectargumentmustbe outside the NP, it is presumablypossible to derive the position of the subject: since it is higher in the argument.structurerepresentation, it must be 0-markedafter every other argument.Therefore, if any argumentis outside the NP, the subject must be also. However, why should an argumenthave to occur outside NP to begin with? In terms of this theory it means that one argumentmust be 0-markedoutside NP, but this makes no sense if argumentsinside and outside NP have the same status. Moreover,why shouldthe subjectnot count as meetingthe requirement, whatever it is? All the argumentsare argumentsof the Verb, so the subject shouldcount in the same way as any other argument. In sum, althoughthe single 0-markeraccount can match the dual 0-markeraccount quite closely in a numberof respects, it does not seem to have the same overall scope. It is importantto emphasizealso that the single 0-markeralternativepreserves many of the essential features that we have been arguingfor here. For example, the Noun and suru would have to be lexically inserted as a pair in this theory too; otherwise, suru could inheritan argumentstructurefrom one Noun and be insertedwith another.Moreover, the explanations sketched above for argumentdistributionin the suru complex depend entirelyon the idea that argumentstructureis hierarchicallyorganized,with the organizationconstraining0-markingin the single 0-markeraccount and Transferin the dual 0-markeraccount. 5. Conclusion The suru complex is formed by a process of complex predicate formation, in which a predicate acquires argumentsthat it does not normallylicense. Complex predicateformation applies to the argumentstructureswith variableresults, sometimes leaving the Noun with no argumentsand sometimes leaving it as an impoverished0-assigner.With this assumption,the exact rangeof possible outcomesfollows fromthe theory of complex predicates and the lexical entries for suru and Nouns. Part of the explanation relies This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 229 crucially on the hypothesis that the light Verb complex contains two predicates. This opens up the possibility that other cases of complex predicate formation might also involve dual 0-markers,ratherthan a single predicate. In causatives and other cases of "clause union," where both predicatesare verbal, or in the Englishlight Verb construction where case markingis not available, it is not always easy to determine where responsibilityfor 0-markinglies. The case-markingsystem of Japanesemakes it easy to monitorthe distributionof argumentsin the light Verb complex. Although the focus of this investigationhas been the phrasal suru complex, there is a widely studied incorporationconstruction, illustratedin (2b) and repeated as (55), in which suru and a Noun form a single compoundword. shita. (55) John-wa Bill-to AISEKI Bill-with John-Top table-sharingsuru-Past A glance at the literatureon Japaneseincorporationcited earlierwill reveal that incorporation exhibits many apparentidiosyncrasies-for example, Nouns that occur in the incorporatedversion, but not in the phrasallight Verb complex. Nevertheless, the productive cases of incorporationcan be straightforwardly understoodin the present terms. They are formed by morphologicalcompounding,appliedto the case of complex predicate formationwhere the Noun yields all its 0-assigningcapacitiesto suru. The argument structure of the output then is a verbalized version of the argumentstructureof the Noun, just as it is in the correspondingphrasalsurucomplex. Whatis being incorporated is the Noun that is lexically listed with the derived argumentstructureof suru. Note that incorporationin Japanese differs from that found in some other languagefamilies (Baker (1985), Rosen (1987)) in that it is not an argument,but part of the complex predicate, that is incorporating.(In fact, there is some evidence (see Grimshaw,Ito, and Mester (in preparation))that what is involved is really incorporationof a Verb, not a Noun.) Of course, the interactionof incorporationandcomplexpredicateformationbears on the correct treatmentof each: if incorporationis lexical (see di Sciullo and Williams (1987), Rosen (1987)), then complex predicateformationitself must have the status of a lexical process, if it is to feed incorporation. The Verb suru illustratesone kind of light Verb: its argumentstructureis so highly underspecifiedthat it is incapableof 0-assignmentof any kind. Other light Verbs, like saseru, have a more fully specified argumentstructure:incomplete, but with some arguments specified. A particularcase of this seems to occur with the Englishlight Verbs in expressions like put the blame on, give someone a kick, take a walk (Higgins(1974), Jackendoff(1974), Oehrle (1975), Wierzbicka(1982), Cattell (1984)). These expressions are almost synonymouswith their verbal counterparts:blame someone, kick someone, and walk, suggestingthat the argumentstructureof blame, kick, and walk carries most of the burden.Nevertheless, the influenceof the Verb itself is detectablein subtle meaning changes. For example, althougha spidercan walk, a spider does not normallytake a walk. This difference presumablyreflects the influence of the argumentstructureof take on the interpretationof the complex. Moreover, there are systematic relations This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 230 JANE GRIMSHAW AND ARMIN MESTER (discussed in the references cited) between the propertiesof the Verbs themselves and those of the Nouns they combine with: give combines with predicatesthat take Goals, for example. The Englishlight Verbs, then, resemblesaseru in havingpartiallyspecified argumentstructures, not the completely empty one of suru. The English facts are extremely complex, in partbecause of the high degree of lexicalizationand in partbecause of the difficultyof determiningthe exact phrasalpositionof the arguments.Nevertheless, it is clear in outline how the theory might apply to them. The Verb provides the subcategorization/casestructurefor the clause level and has an incompleteargumentstructure, whereas the Noun has a complete one. Argumentsof the Noun transfer to the Verb, as for suru. Where the Verb already has a correspondingargument,there are a number of possibilities, the most likely being that the two argumentpositions simply merge into one. Whether the general principles governing suru light Verb complexes can be detected in the English counterpartsremainsa question for investigation,but in outline, at least, this resembles proposals made for English (see Cattell (1984)). Comparisonof suru and the English light Verbs raises another question. Some of the English light Verbs are ditransitiveand assign case to two object NPs. Examples include give someone a kick, make someone an offer. In these cases it is the second NP that is transparent,not the first. So objects of transitive Verbs and second objects of ditransitiveVerbs can be transparent.Why do subjects and first objects of ditransitives not participate?Presumablythis is related to the degree of semantic cohesion between the Verb and its arguments:it has often been noted that idioms seem to treat the Verb and second object as a unit, leaving the first NP position free: give NP the slip, for example. Moreover, idioms involving Verbs and their external argumentsare certainly rare if not impossible. Eventually, then, the generalizationabout which NPs can be transparentshould reduce to the theory of possible complex lexical entries. Two hypotheses about argumentstructurerepresentationare crucially invoked in this study. The first is the idea that Nouns do not have subject arguments, which is central to the account of why the outside occurrenceof the subject is not sufficientfor a well-formed complex. The second is the hypothesis that the list of argumentsin an argumentstructureis hierarchicallystructured.This lies behind the inside/outsidedistributionof internalarguments,as well as the fact that the subject of the complex must be outside NP. Whatthen determinesthe hierarchicalstructure?The proposaldeveloped in Grimshaw(to appear) is that a thematic hierarchy(like that of Jackendoff (1972)) determinesthe organizationof argumentstructure.The centralpropertiesof light Verb complexes with suru then follow from the theory of argumentstructureand complex predicate formation. References Anderson, M. (1983-4) "PrenominalGenitive NPs," The LinguisticReview 3, 1-24. Aronoff, M. (1976) WordFormationin GenerativeGrammar,MIT Press, Cambridge,Massachusetts. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions LIGHT VERBS AND 0-MARKING 231 Baker, M. (1985)Incorporation:A Theoryof GrammaticalFunction Changing,Doctoral dissertation, MIT, Cambridge,Massachusetts. Bresnan, J. (1982) "Controland Complementation,"LinguisticInquiry13, 343-434. Burzio, L. (1981) Intransitive Verbs and Italian Auxiliaries, Doctoral dissertation, MIT, Cambridge, Massachusetts. Cattell, R. (1984)CompositePredicates in English, AcademicPress Australia,North Ryde, New South Wales. Chomsky, N. (1981)Lectures on Governmentand Binding, Foris, Dordrecht. Chomsky,N. (1982)Some Conceptsand Consequencesof the Theoryof Governmentand Binding, MIT Press, Cambridge,Massachusetts. Chomsky, N. (1986)Barriers,MIT Press, Cambridge,Massachusetts. di Sciullo, A. M. and E. Williams (1987) On the Definition of Word, MIT Press, Cambridge, Massachusetts. Grimshaw,J. (1986) "Nouns, Argumentsand Adjuncts," ms., Linguisticsand CognitiveScience Program,BrandeisUniversity, Waltham,Massachusetts. Grimshaw,J. (to appear)ArgumentStructure,MIT Press, Cambridge,Massachusetts. Grimshaw,J., J. Ito, andA. Mester(inpreparation)"Nouns, Verbs, andJapaneseIncorporation." Hale, K. (1983) "Warlpiriand the Grammarof Non-configurationalLanguages," Natural Language and Linguistic Theory1, 5-47. Harada, S. (1973) "Counter Equi NP Deletion," Annual Bulletin 7, Research Institute of Logopedics and Phoniatrics,University of Tokyo. Hasegawa, N. (1979)"A Comparisonof Two Approachesto a Class of JapaneseConstructions," ms., University of Washington,Seattle. Higgins, F. R. (1974)ThePseudo-cleftConstructionin English, Doctoraldissertation,MIT, Cambridge, Massachusetts. Inoue, K. (1976)HenkeiBunpoo to Nihongo [Transformational Grammarand the JapaneseLanguage], Vol. 1, Taishukan,Tokyo. Jackendoff,R. (1972) Semantic Interpretationin GenerativeGrammar,MIT Press, Cambridge, Massachusetts. Jackendoff,R. (1974) "A Deep StructureProjectionRule," LinguisticInquiry5, 481-505. Jespersen,0. (1954)A ModernEnglishGrammaron HistoricalPrinciples,GeorgeAllen & Unwin, London, and EjnarMunksgaard,Copenhagen. Kageyama, T. (1976-77) "Incorporationand Sino-JapaneseVerbs," Papers in Japanese Linguistics 5, 117-156. Kageyama,T. (1982) "WordFormationin Japanese," Lingua 57, 215-258. Kayne, R. S. (1981) "ECP Extensions," LinguisticInquiry12, 93-133. Kuroda, S.-Y. (1965)GenerativeStudies in the Japanese Language, Doctoral dissertation,MIT, Cambridge,Massachusetts. Kuroda, S.-Y. (1978) "Case-Marking,CanonicalSentence Patterns, and Counter-Equiin Japanese," in J. Hinds and I. Howard, eds., Problems in Japanese Syntax and Semantics, Kaitakusha,Tokyo. Levin, B. and M. Rappaport(1986) "The Formationof AdjectivalPassives," LinguisticInquiry 17, 623-661. Oehrle, R. T. (1975) The GrammaticalStatus of the English Dative Alternation, Doctoral dissertation, MIT, Cambridge,Massachusetts. Perlmutter,D. (1978) "ImpersonalPassives and the Unaccusative Hypothesis," in Proceedings of the Fourth Annual Meeting of the BerkeleyLinguistics Society, UCLA, Los Angeles, California. This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions 232 JANE GRIMSHAW AND ARMIN MESTER Poser, W. (1980) Untitled manuscript,Departmentof Linguistics and Philosophy, MIT, Cambridge, Massachusetts. Rappaport,M. (1983) "On the Nature of Derived Nominals," in L. Levin, M. Rappaport,and A. Zaenen, eds., Papers in Lexical Functional Grammar, Indiana University Linguistics Club, Bloomington. Rosen, S. (1987) "Two Types of Noun Incorporation:A Lexical Analysis," ms., BrandeisUniversity, Waltham,Massachusetts. Safir, K. (1987) "The Syntactic Projectionof Lexical Thematic Structure,"Natural Language and Linguistic Theory 5, 561-601. Saito, M. (1985) Some Asymmetries in Japanese and Their Theoretical Implications, Doctoral dissertation,MIT, Cambridge,Massachusetts. Shibatani,M. (1973)"Semanticsof JapaneseCausativization,"Foundationsof Language 9, 327373. Stowell, T. (1981) Originsof Phrase Structure,Doctoral dissertation,MIT, Cambridge,Massachusetts. Wierzbicka, A. (1982) "Why Can You Have a Drink When You Can't Have an Eat?" Language 58, 753-799. Williams, E. (1981) "ArgumentStructureand Morphology,"TheLinguisticReview 1, 81-114. Zubizarreta,M. L. (1985)"The RelationbetweenMorphophonologyand Morphosyntax:The Case of RomanceCausatives,"LinguisticInquiry16, 247-289. (Grimshaw) Linguistics and Cognitive Science Program Department of Psychology Brandeis University Waltham, Massachusetts 02254 (Mester) Department of Linguistics Calhoun Hall 501 University of Texas at Austin Austin, Texas 78712 This content downloaded from 141.164.71.238 on Tue, 3 Dec 2013 11:00:45 AM All use subject to JSTOR Terms and Conditions