Features, Syntax, and Categories in the Latin Perfect Davidembick

Features, Syntax, and Categories in the Latin Perfect DavidEmbick

Theanalysis centers on thenotion of category in synthetic and analytic verbalforms and on thestatus of thefeature that determines the forms ofthe Latin perfect. In this part of the Latin verbal system, active formsare synthetic (‘ ‘verbs’’) butpassive forms are analytic (i.e., participleand finite auxiliary). I showthat the two perfects occur in essentiallythe same structure and are distinguished by adifferencein movementto T; moreover,the difference in forms can be derived withoutreference to category labels like ‘ ‘Verb’’ or‘ ‘Adjective’’ on theRoot. In addition, the difference in perfects is determined by a featurewith clear syntactic consequences, which must be associated arbitrarilywith certain Roots, the deponentverbs.I discussthe implica- tionsof these points in the context of Distributed Morphology, the theoryin whichthe analysis is framed.

Keywords: syntax/morphologyinterface, category, features, passive voice,Distributed Morphology

1Introduction Questionssurrounding the relationship between syntactic and morphological definitions of cate- goryhave played and continue to play an importantrole in grammatical theory. Similarly, issues concerningthe type, nature, and distribution of features in different modules of the grammar definea numberof questions in linguistic theory. In this article I examinethe syntactic and morphologicalprocesses and features at playin theconstruction of analyticand synthetic verbal forms, andin the determination of differentsurface categories. I focusprimarily on thefact that theLatin perfect is syntheticin the active voice (e.g., ama¯v¯õ ‘I(have)loved’ ) butanalytic in the passive,with a participialform ofthemain verb and a form oftheauxiliary ‘ be’( ama¯tus sum). Theoretically,the analysis addresses (a) thestatus of category in syntax and morphology, and (b) thestatus of thefeature underlying the analytic /syntheticdifference. Beginningwith category, the notion ‘ ‘participle,’’ whichplays a centralrole in the discussion ofanalyticverb forms, hasa dualstatus: on one level defined morphologically (‘ ‘verbthat agrees

Iwouldlike to thankMorris Halle, Alec Marantz,and Rolf Noyer for numerous detailed discussions of thismaterial. Inaddition, I havebenefited from comments at variousstages fromElena Anagnostopoulou, Karlos Arregi,Rajesh Bhatt, JonathanBobaljik, Noam Chomsky,Paul Elbourne, Martin Hackl, Sabine Iatridou, Liina Pylkka ¨nen,Norvin Richards, DonRinge, and Philippe Schlenker. Parts ofthisarticle were alsopresented at theSecond Penn /MITWorkshop on the Lexicon,and I wouldlike to thank the workshop participants and members ofthe audience for additional comments. Finally,I wouldlike to thank the JST /MIT[Mind Articulation] Project for support during the time thatthe research presentedhere was conducted.

Linguistic Inquiry, Volume 31, Number 2,Spring 2000 185–230 q 2000 bythe Massachusetts Institute ofTechnology 185 186 DAVIDEMBICK inthe ‘ adjectival’pattern’ ’), onanother level defined in terms of syntactic distribution (‘ ‘ verb appearing(in certain contexts) with a finiteauxiliary and with certain morphosyntactic features, etc.’’). Underlyingthese types of definitions is the intuition that a participlemay be defined as averbbehaving in some sense nonverbally. In traditional terms, this is akin to the classification ofa participleas a typeof (or asrelated to) a ‘‘deverbaladjective,’ ’ thatis, as part nominal (4 adjectival),part verbal. 1 Classificationsof this type lead naturally to the positionthat participles arein somesense ‘ ‘derived’’ asopposed to ‘ ‘primitive,’’ butdo notilluminate the nature of the derivation. Inthe abstract, we mayconsider two possible types of analysis in whicha V(erb) isrealized asaparticiple.The first sees an underlying V asbeingof necessityconverted into the category A(djective).On this view the designations V and A havesubstantive content, in that they are associatedwith both syntacticosemantic and morphological effects. For instance,the morphologi- calproperties of the derived A differfrom thoseof the original V, inthat it would show gender/number/casedistinctions; at thesame time, properties such as Case assignment might also bedifferent in the derived A thanin the original V. Whateverthe particulars of the category- changingoperation are (i.e., whether it takes place in a lexicon,or by virtue of a syntacticAP dominatinga VP), theresult is the same: a clausecontaining the participle is distinct from one inwhich ‘ ‘regular’’ verbsare found. Thesecond alternative is that the syntax of participlesdoes not involve an ‘ ‘adjectivalizing’’ projectionAP aboveVP, orany lexical operation; instead, the category change from VtoA is onlyrelevant after the syntactic derivation. This is thetype of approachthat I develophere. The basisfor thisis an argument showing that the same syntactic structure underlies both auxiliary ` participleformations and finite verbs in the perfect. That is, the two types of perfect appear inthesame syntactic structure and differ only in the position to which the verb has raised in the tree.This treatment has clear implications for thenotion of category: different morphological categoriesare realized in thetwo perfects, but the syntactic structure from whichthey are derived isthesame. Thus, the difference in morphologicalcategory does not correspond to a difference insyntacticcategory. Against this background, the syntactic distribution and morphologicalbehav- iorof ‘‘pastpassive participles’ ’ foundin the perfect is shown to followdirectly from thesyntactic derivationand independently required properties of Latin morphology. No category-changing syntacticposition is needed to derive the participial form —itsappearance and behavior follow from generalprocedures operating on aspecificsyntactic structure. Rather than being ‘ ‘adjectives derivedfrom verbs,’’ participlesresult from therealizationof label-neutralRoots in aconfiguration thathas components associated with finite verbs, but that differ in crucial respects to be made explicitbelow. 2

1 Thus,for instance, in classical grammar, thesource of the Latin term participium is tobe found in the fact that participlesbehave in part like verbs, expressing temporal /aspectual distinctions,and in part like nouns, varying for case. See Varro 1938:VIII58 /X 17. 2 Theoriesemploying Roots underspecified in this manner are advancedin Marantz 1995,1997, developing ideas fromChomsky 1970; additional details of thisapproach will be made clear below. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 187

Thearguments I developare presented in thecontext of aparticularconception of modularity inthe grammar. In order to make this clear, I willfirst reviewsome basic properties of Distributed Morphology(Halle and Marantz 1993 and related work), the theory in which the analysis is framed.Distributed Morphology operates in terms of LateInsertion, theidea that phonological piecesinstantiate terminals containing abstract features postsyntactically, with the syntax proper manipulatingsets of features. 3 Iwillrefer totheactual pieces of phonologicalmaterial as exponents. Theseexponents, along with a statementof the features they instantiate, are vocabulary items. Particularvocabulary items compete for insertioninto morphosyntactic positions, with the itemmost highly specified for thefeatures on a nodetaking precedence over less-specified items; thisis disjunctive realization, as in Anderson 1986, 1992. Defined in terms of positions,this is inaccordance with the Subset Principle, such that the vocabulary item specified for thegreatest subsetof featureson a terminalwill be inserted into that position. (1) SubsetPrinciple Thephonological exponent of avocabularyitem is insertedinto a positionif theitem matchesall or a subsetof the features specified in the terminal morpheme. Insertion doesnot take place if thevocabulary item contains features not presentin themorpheme. Whereseveral vocabulary items meet the conditions for insertion,the item matching thegreatest number of features specified in the terminal morpheme must be chosen. (Halle1997:427) Afurtherbackground assumption concerns the distinction between the functional and lexical vocabulariesof a language.I willassume that functional categories merely instantiate sets of abstractsyntacticosemantic features. Thus, notions like ‘ ‘determiner,’’ ‘‘tense,’’ andso on, are definablein terms of these features. The open-class (lexical) vocabulary items, referred toas Roots, arenot simply the realizations of featurebundles. They consist of phonologicalrepresenta- tions,which have encyclopedic (i.e., not purely featural) meanings. 4 Whatother content they possess—for instance,whether they are specified for syntacticcategory or for semanticfeatures—isprecisely what is atissue. Accordingto a furtherhypothesis, expressed in Marantz 1994, 1995, Roots too are subject toLate Insertion; that is, they are only inserted into syntactic structures in the morphological component,like functional vocabulary items. 5 Thisposition on modularity makes specific predic- tionsabout the flow of information between components of the grammar. Specifically, in the simplestcase arbitrary features of vocabulary items could not affect the syntax, because the

3 Theposition that morphosyntactic and morphophonological features are distinctfrom one another is notunique toDistributed Morphology; it is theinstantiation of the SeparationHypothesis (named byBeard (1966))and is assumed ina numberof otherframeworks. Differences between DistributedMorphology and other separationist theories will be takenup in section 8. 4 See Harley andNoyer 1998 for a discussionof some ofthe consequences of the differences invocabulary types. 5 Or,positions that have been added to this structure in the morphology; see below. 188 DAVIDEMBICK vocabularyitems are not present in the syntax. With its converse, this position is stated in the notion of FeatureDisjointness, definedas follows: (2) FeatureDisjointness Featuresthat are phonological, or purelymorphological, or arbitraryproperties of vocabularyitems, are not presentin the syntax;syntacticosemantic features are not inserted inmorphology. Thisposition is a clearconsequence of thehypothesis that Late Insertion is universal, that is,applies in all possible cases, to functional morphemes and Roots alike. The resulting view of featuresis based on the following reasoning. Particular Roots and functional vocabulary items arenot present in the syntactic computation. This has the effect of imposing a distinctionamong thetypes of features that are found in syntax and morphology. Feature Disjointness is a strong hypothesisabout how features in the grammar willinteract; and, to the extent that it is valid, it followsdirectly from thestructure of thegrammar outlinedabove. In effect, it makes a particularly strongclaim concerning the hypothesis of LateInsertion. Theaspect of thisposition on features that is relevant for thepresent discussion is theidea thatfeatures that are the idiosyncratic properties of particular Roots are not present in thesyntactic derivation.Conjugation and declension class features, which are simply memorized with particular nounsor verbs,are clear examples of featuresof thiskind; although they are required for morphologicalwell-formedness in languages like Latin, they have no syntacticstatus. Similarly, phonologicalproperties of Roots(e.g., ‘ ‘beginswith /s/-’’ )seemto be irrelevant for thesyntax. Thetype of argumentthat can be madein thisframework isthen straightforward. If itcan beshown that a featureis an arbitrary property of a Root,and that the same feature figures cruciallyin someprocess, then the conclusion that should follow is that the feature and the process inquestionare morphological (i.e., not presentin thesyntactic computation). In the caseI examine here,however, the feature found to beresponsiblefor theforms ofthe Latin perfect is shownto be(a) anarbitraryproperty of certainRoots, and (b) afeaturethat figures crucially in a syntactic process.The implication is thatthe feature in questionhas to bepresent in thesyntax, and at the sametime it must be associated with certain Roots inherently. This aspect of the analysis has clearconsequences for thetheory of features and their distribution in different modules of the grammar.In addition, it raisesserious questions about the claim that Roots are not present in the syntacticderivation. After consideringtwo analyses of the feature in question that are consistent withthe idea that Roots are always inserted postsyntactically, and examining their shortcomings, Ioffer ananalysis suggesting that certain Roots may be present in the syntactic derivation to beginwith, and I examinethe issues surrounding the various treatments of this feature. After clarifyingthe status of thefeature involved in the Latin perfect, I explorethe implica- tionsthe analysis has for notionsof category,and the role of categoryin syntaxand morphology. Theanalysis I presentis onein whichcategory labels for Roots(i.e., for thelexical vocabulary) aredispensed with entirely. The surface morphological differences between ‘ ‘verbs’’ and‘ ‘adjectives’’ reducesto the syntactic structures in whichRoots appear, in conjunctionwith agreement FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 189 processesdefined to operate in termsof these structures. The most important role in determining what‘ ‘category’’ aRootis realizedas is played by thefunctional heads in the local environment oftheRoot. These consist of abstract morphosyntactic features and are labeled as D(eterminer), T(ense),Asp(ect), and so on. The implications of thistreatment, and a comparisonof theanalysis presentedhere with other models of syntax /morphologyinteractions, are presented in the conclud- ingsections.

2TheLatin Perfect Thebasic facts to be investigated are as follows. Latin shows a splitin the realization of its perfecttense /aspectforms. 6 Theperfect active is asyntheticform withan aspectual affix and a specificset of endings varying for person /number,as shown in (3). 7 Theperfect passive, on the otherhand, is analytic,with a participleagreeing in genderand number, and the verb ‘ tobe’ (4). 8

6 Thesame is trueof theother parts of theperfect system, thatis, the pluperfect and the future perfect. Interms ofitsinterpretation, the forms called ‘‘perfect’’ inLatin are eitherperfect orperfective.This is aresultof thefact thatthe Latin perfect represents thecollapse of twodistinct Indo-European categories, the perfect andthe aorist, onbothmorphological and semantic levels.The ambiguity of theperfect formis notedby later Latingrammarians, for example,Priscian in the 5th century (see Wackernagel1920, Binnick 1991): ‘ ‘We havethe perfect forboth the Greek perfect andGreek aorist’’ (fromWackernagel 1920:187). Relating the Latin situation to Greek doesnot clarify the matter entirely,as theGreek aoristis notalways interpretedas a‘‘pastperfective.’ ’ Moreover,the interpretation of the Classical Greek perfect isitself somewhat complex;it is elucidated in thestudy of Wackernagel (1904) (see alsoChantraine 1927 and,for a recent compilation /analysis,Sicking and Stork 1996). For a discussionof some oftheLatin facts, see Serbat 1980and other papers in that volume. Theoretically, this behavior in the perfect raises anumberof questionsabout how thefeatures accountingfor these twointerpretations are distributedin the clause. As Icannothope to address these here, Iwillsimply use thefeature [perf]as anabbreviation for relevant syntacticosemantic features. One furthercase is worthnoting. In some cases thepast participle with ‘ be’has astativeinterpretation, as inthe Englishstative passive Thedie is cast. Insuch cases, althoughthe surface formis thesame as thatfound with the passive perfect, theunderlying structure is presumablyone in which ‘ be’takes asmall clause complement,or in any case a structurethat differs from the ‘ ‘normal’’ clauses studiedhere. 7 There are anumberof morphologicalforms associated withthe perfect active,stemming, as notedabove, from thefact thatthis category represents thefusion of twodistinct categories inIndo-European.The following set offorms fromAllen 1931 illustrates some ofthevariety found in the formation of theperfect: (i)a. -v-:amo ¯‘love’, perf.am-a ¯-v- ¯õ b.-v- on athematic form:moneo ¯ ‘warn’, perf.mon-u- ¯õ c. -s-: scribo¯‘write’, perf.scrip-s- ¯õ (cf. Greek s-aorist) d.Reduplication: cado ¯ ‘fall’, perf.ce-cid- ¯õ (cf. Indo-Europeanperfect) e. Vowel change:video ¯‘see’, perf.v ¯õ d-¯õ f.No change: verto ¯‘turn’, perf.vert- ¯õ Thelast three typeshere havea À-allomorphof theperfect suffix,along with (in some cases) morphophonological readjustmentto the stem. Inaddition, some ofthe person /numberendings found in the perfect are uniqueto this category (e.g., the 1sg. and 2sg.);others contain familiar componentsin additionto pieces foundonly in the perfect. 8 Inpresenting the endings - ¯õ, -ist¯õ, andso on, in the active, I haveleft an- is-component,found in the 2sg., 2pl., and3pl. (at least historically;it is changedby rhotacismto - er-inthe last case), unsegmented;it is occasionallyregarded as anindependent formant. This is diachronically the same as the- is-foundin the pluperfect, discussed in footnote 49. 190 DAVIDEMBICK

(3) Perfectindicative active 9 (4) Perfectindicative passive 1sg.am-a ¯-v- ¯õ 1sg.am-a ¯-t-us /-a/-um sum ‘Ihaveloved’ ‘I was/havebeen loved’ 2sg. am-a¯-v-ist¯õ 2sg. am-a¯-t-us/-a/-um es ‘youhave loved’ ‘you were/havebeen loved’ 3sg.am-a ¯-v-it 3sg.am-a ¯-t-us /-a/-um est ‘he/shehas loved’ ‘he/she was/hasbeen loved’ 1pl.am-a ¯-v-imus 1pl.am-a ¯-t- ¯õ /-ae/-a sumus ‘we haveloved’ ‘we were/havebeen loved’ 2pl. am-a¯-v-istis 2pl. am-a¯-t-¯õ /-ae/-a estis ‘youhave loved’ ‘you were/havebeen loved’ 3pl. am-a¯-v-e¯runt 3pl. am-a¯-t-¯õ /-ae/-a sunt ‘theyhave loved’ ‘they were/havebeen loved’ Oneresponse to this situation as it ispresentedin descriptivegrammars isto regard it asa matterof classificatory importance only, without a basisin the linguistic system. For instance, onemight hold that the analytic passive perfect is not the same as the synthetic perfect in terms ofthetemporal /aspectualfeatures involved, but is instead something composed of different parts toachieve a similarmeaning. Alternatively, the two could be the same in terms of the features involved,but the features would be distributed differently in the hierarchical structure. If either ofthese approaches were correct,the difference in form wouldhave to be ultimately reducible toadifferencebetween passive and active syntax. In the next section I showthat this is notthe case.In terms of its morphosyntactic composition (i.e., at the level of features distributed in a syntacticstructure), the analytic perfect is just as much a partof the ‘ ‘verbal’’ systemas any syntheticform. 1 0 Therealization of the Root as a participlefollows from propertiesof Latin morphology.

3Syntaxand Morphology in the Perfect Thearguments of thissection show that the appearance of analytic versus synthetic forms inthe Latinperfect is determinedby the presence of a feature[pass], which is (a) systematicallycorre- latedwith the syntax of passivization,and (b) inherentlypossessed by certainRoots, for reasons thatare not related to passive syntax. As notedin the preceding section, one assumption might be that active and passive syntax aresimply different from eachother in theLatin perfect, such that the passive is, by virtue of its

9 Inexamples andtables Iwillrepresent Latinorthographically. 10 Traditionalanalyses ofLatin have also taken the position that the analytic perfect is actuallypart of the verbal system onthe same levelas thesynthetic forms; see, forinstance, Brugmann 1895. The reasoning in such cases was basedon thefact thatthe interpretation of ‘ be’`participlewas distinctfrom that of ‘be’and a ‘‘simple’’ adjective;the former hadthe temporal and aspectual propertiesnoted in footnote 6, whereas thelatter was simplypresent tense. For furtherdiscussion of the transition from ‘ ‘deverbaladjective’ ’ to‘ ‘participle’’ intraditional terms, see Wackernagel1920: 288ff. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 191 syntax,necessarily analytic. 1 1 Evidenceagainst this view is provided by the perfect forms of certain deponent verbs.Deponent verbs are only capable of appearingin passive form, butmay appearin active syntax, as Iwillshow below. 1 2 For illustration,compare the present passive forms of amo¯ ‘love’(a regularverb) with the presentforms of hortor ‘toexhort’ (a deponent). (5) Presentpassive (6) Presentof deponent am-or ‘Iamloved’ hort-or‘ Iexhort’ am-a¯-ris ‘youare loved’ hort-a¯-ris ‘youexhort’ am-a¯-tur ‘he/sheis loved’ hort-a¯-tur ‘he/sheexhorts’ am-a¯-mur ‘we areloved’ hort-a¯-mur ‘we exhort’ am-a¯-min¯õ ‘youare loved’ hort-a¯-min¯õ ‘youexhort’ am-a-ntur‘ theyare loved’ hort-a-ntur‘ theyexhort’ Inaddition to the forms in(5), normal verbs like amo¯ haveactive forms. For thepresent indicative,these are as follows; these active forms simplyare not found with deponents like hortor: (7) Presentindicative active am-o¯ ‘I love’ am-a¯-s ‘youlove’ am-a-t ‘he/she loves’ am-a¯-mus ‘we love’ am-a¯-tis‘ youlove’ am-a-nt‘ theylove’ Theidentity between deponent verbs and the passives of normalverbs extends to theperfect. Deponentverbs have no synthetic perfect forms, butare instead always analytic in the perfect. (8) Perfectof a deponent *hort-a¯-v- ¯õ ‘I(have)exhorted’ ; compare(3) hort-a¯-t-ussum ‘ I(have)exhorted’ Thedescription above covers the basic morphological properties of deponent verbs. I now showthat certain deponent verbs— specifically, a certainset of transitives— must be treated as

11 Thispossibility has tobe stated as specific tothe perfect, giventhat other passives inLatin are synthetic (cf. laud-o¯ ‘I praise’, laud-or ‘Iam praised’). 12 Deponentverbs are coveredin detail in all major descriptivegrammars andhandbooks of Latin.Specific studies thatbear onthe issues addressedhere willbe cited as thediscussion proceeds. I am referringto the forms ofdeponents andverbs in passive syntax as ‘‘passive’’ onlyfor convenience, and in accordance withstandard usage for Latin. As I willmake clear below,this form is notexclusively associated withpassive syntax, but also appears withsyntactically related formations. Inaddition to theverbs that appear as nonactivein all tenses andmoods, there are afew verbsthat show different voiceforms indifferent tenses. Verbs thatare deponentonly in the perfect are referred toas ‘‘semideponent’’ (cf. audeo¯ ‘I dare’, ausus sum ‘Ihavedared’ ); otherverbs of this type are fideo¯ ‘trust’, gaudeo¯ ‘rejoice’, and soleo¯ ‘be wont’. Sommer (1914)notes that many of these verbshad synthetic, active perfects inearly Latin. 192 DAVIDEMBICK beinginherently specified for afeatureresponsible for theirinflectional properties, which I label [pass].1 3 Syntactically,some verbs of this type are not different from theiractive, nondeponent counterparts. 1 4 Effectively,then, certain verbs possess this feature for reasonsthat have nothing todo with passive syntax or relatedsemantics. If theverbs in question appeared in passiveform for systematicsyntacticosemantic reasons, theywould be expectedto form anaturalclass; but they do not.The conclusion that there is no commonsyntactic or semanticbasis for unitingthe deponent verbs is sharedby earlier authors; seeDraeger 1878:149and Meillet 1966, along with Baldi 1976 and references found there. This approach,holding that synchronically deponents have their special property somewhat arbitrarily, iscontrastedwith other traditional treatments of thetopic, which seek a lexicosemanticbasis for the class.1 5 Thelexicosemantic approach also defines more recent typological studies of voice (e.g.,Klaiman 1991, Kemmer 1993,1994, and related work). Before I proceedto the syntactic propertiesof transitive deponents, I willaddress points raised by the ‘ ‘semantic’’ approachto deponentverbs, concentrating on Kemmer’ s (1993)discussion. Kemmer’s functional /typologicalaccount of the ‘ ‘middlevoice’ ’ (includingin thiscase the Latin‘ ‘passive’’) takes middlemorphology asaunifiedsemantic concept, expressing the limited differentiationof participants (whether actors or events) in thesituation described. Thus, in comparisonwith normal, active form, themiddle ( 4 passivein Latin) forms indicatethat actors or eventsare in some sense undifferentiated. Discussions of this type are difficult to assess in the presentcontext, in thatthey make no specificclaims about how syntax, semantics, and morphology interactwith one another;the emphasis is instead on correlating morphological forms withsituation types.Without getting into the details of how such a semanticapproach to voice fares more generally,the specific point to be made here with respect to Latin is that this type of approach hasfailed to offer independentreasons for thinkingof certaintransitive deponents as beingsyntac- ticosemanticallydistinct from regulartransitives. To the extent that Kemmer’ s discussionmakes anypredictions about the syntax at all,it seemsto predict that deponents should be ‘‘lesstransi- tive’’ thantheir nondeponent counterparts. However, no criterion independent of the ‘ ‘middle’’ morphologyitself is putforth as a meansof identifying this ‘ ‘middle’’ semantics. 1 6 This is not

13 Itmight be thecase thatthere are unergativedeponents as well; butI willfocus on the transitives here. 14 Otherverbs classified as deponentshow passive forms forreasons thatare arguablysyntactic. For instance, a numberof intransitive deponents, such as morior ‘die’, can betreated as unaccusatives(Embick 1997) and fall into identifiablelexicosemantic classes (cf. Levinand Rappaport-Hovav 1995). I willfocus here ontheset forwhich such a syntacticanalysis is not possible. 15 Theapproach seeking a unifiedsemantics alsohas alonghistory, and is closelytied to the question of whether there is aunifiedsemantic basis forthe Indo-European middle voice. For a summary oftraditional approaches to the latter,see, forexample, Gonda 1960a,b. The typical strategy in such treatments isto (a) attempt toidentifya Gesamtbedeu- tung (‘unifiedmeaning’ ) forthe ‘ ‘middle’’ and(b) to simply assert thatall deponentsare somehowconnected with this. Iwillnot attempt toreview theliterature here, but even for Latin alone it is vast;No ¨lting1859 already contains a review ofmuch earlier discussion.Baldi (1976) documents the fact thatalthough certain subregularitiesmay befound with deponents,there is nosingle semantic categorythat unifies them. 16 Partof Kemmer’s argumentfor there beinga semantic correlate todeponent verbs is typological;she identifies anumberof cases inwhich verbs of thesame meaningappear as deponentin unrelatedlanguages. These correlationsseem tostem primarilyfrom the fact thata numberof the verbs classified as deponentin different languages are unaccusativesor psychpredicates, that is, classes inwhich a certain amountof crosslinguistic semantic similarityobtains; see footnote 14. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 193 surprising,for asI willnow show,there are certain transitive deponents that behave indistinguisha- blyfrom normalagentive, transitive verbs, beyond possessing [pass] inherently. Thearguments for treatingcertain deponents as possessing[pass] inherentlyare as follows. Tobegin with, in terms of their basic appearance, transitive deponent verbs behave like other (i.e.,nondeponent) transitives; that is, they take nominative subjects and accusative objects. 1 7 (9) Puer m¯õ litemsequi-tur. boy-NOM soldier-ACC follow-PASS.3SG ‘Theboy is following the soldier.’ Thereis thusno reason to suspectthat the syntax here is anything other than that found with normaltransitives. Specifically, there is no reasonto suppose that these verbs are really ‘ ‘covertly passive,’’ thatis, that they involve syntactic operations related to passivization. 1 8 Thereis, moreover,no reasonbased on lexical semantics to treatcertain transitive deponents as anything other thannormal transitives. This can be seenhere with co¯nsector ‘huntdown’ and aggredior ‘attack’ (alsosee (27a)). (10)a. L. LiciniusCrassus co ¯nsul quo¯sdam . . . co¯nsecta¯tus est L.LiciniusCrassus consul PRON-ACC.INDEF ...hunt-down- PART be-PRES.3SG et interfe¯cit. and kill-PERF.3SG ‘L.LiciniusCrassus when consul hunted down and destroyed a certaingroup . ..’ (Cic., De¯Inventione II.32.111) b.Cethegus Cicero ¯nis ia¯nuam obside¯ret eumque CethegusCicero- GEN door-ACC beset-IMPERF-SUBJ -3SG him-ACC4and v¯õ aggredere¯tur. violentlyattack- IMPERF-SUBJ -3SG ‘Cetheguswas tobeset Cicero’ s doorand assault him.’ (Sall., Cat. XLIII.2) Thepattern exhibited above is telling, as it removes one option for theanalysis of these verbs.One way of implementingthe proposal that deponents really are ‘ ‘passive’’ insomesense wouldbe to say that all deponents are syntactically distinct from agentivetransitive verbs. A plausibleoption at the outset would be to hold that the deponents are like certain psych verbs, withderived subjects and thus passive morphology. 1 9 Verbs ofthis type would be expected to

17 Asmall set oftransitive deponents ( vescor ‘eat’, u¯tor ‘use’, anda few others)takes ablativeobjects, at least in Classical Latin;Gildersleeve andLodge (1895) note that earlier these verbstook the accusative. Thequestion of case withpassives anddeponents is potentiallyof interestbecause ithas beenclaimed that‘ ‘passives’’ sometimes havethe abilityto assign accusative case; see Sobin1985. I willnot address thisissue here,but will instead present a series of argumentsshowing that there is noreason to believe that verbs of therelevant type are actuallypassive. 18 See section5 fora discussionof whatsyntactic configurations are related tothe [pass] feature. 19 Forderived subjects with psych verbs, see analyses alongthe lines pursued in Bellettiand Rizzi 1988and subsequent work;the idea that‘ ‘passive’’ morphologyappears instructureswithout external arguments is discussedbelow. 194 DAVIDEMBICK havethe semantics and syntactic behavior of psych verbs crosslinguistically. But this is not the casewith verbs like aggredior and co¯nsector, whichare deponent and do notact like psych verbs. Theseverbs are as transitive,active, and agentive as possible. Afurtherfact fits in with this treatment of certaindeponents as being normal transitive verbs syntactically.In addition to appearing in transitivesyntactic environments, some deponents like hortor mayalso appear in passive syntax; thus, consider the following example, in which the verbin passive syntax is morphologicallyidentical to the one that appears in active syntax: 2 0 (11) Ab am¯õ c¯õ shorta¯-re¯-tur. byfriends urge- IMPERF.SUBJ-PASS .3SG ‘He was urgedby friends.’ (subjunctive) (Varro inPrisc., GL II 387,2) Theexistence of such verbs provides a strongargument that these verbs are syntactically likeother active transitive verbs: that is, they are so transitive that they are capable of being passivized. Finally,there are agentive nominalizations formed from therelevant deponent verbs: for sequor,sec-u ¯-tor ‘pursuer’; for aggredior,aggres-sor ‘attacker,assailant’ ; andfor co¯nsector, co¯nsecta¯-trix ‘onewho pursues or strivesafter’ (fem.) (Glare 1982). Thecases adduced above illustrate clearly the point that passive syntax, and the features underlyingit, is distinctfrom thefeature [pass] thatresults in passiveforms. Thatis, the feature underlyingpassive forms maybe presentindependently of passive syntax. I concludefrom these argumentsthat the relevant deponent verbs are specified for thefeature [pass], which is simply apropertyof the Root. The exact details of thisspecification will become clear in section 5. Tosummarize the discussion of deponents to this point: The feature [pass] isin this case simplya propertyof particularRoots, which, when inherently possessed like this, is systematically unrelatedto passive syntax. In this way it has a statussimilar to that of conjugation features, whichare crucial for morphologicalwell-formedness. The difference between the two is thatthe [pass] featureis sometimes inherently possessed and sometimes systematically correlated with passivesyntax (see below). Recallnow that deponents and verbs in passive syntax behave identically for thepurposes ofinflection and for thepurposes of the formation of the perfect. This identity is captured as follows.The [pass] featurethat is foundin syntacticpassives is associatedwith syntactic configura- tionsin which there is no external argument (see Embick1997, 1998 for arecenttreatment of

20 Verbs ofthis type, with only ‘ ‘passive’’ formbut with both passive and active syntax,are referred toas verba commu¯nia. Accordingto the Latin author Aulus Gellius (2nd century), the Latin grammarians usedthis term because the verbsin question had one form (the nonactive) that was common toboth active andpassive interpretations (see Gellius 1927).This classification is alsofound in extant Latin grammars, whichpostdate Gellius. The commu¯nia comprise one ofthefive voices ( genera verb¯õ, literallysomething like ‘ ‘verbalgenders’ ’) recognizedin theLatin grammatical tradition. Foran overview discussion of thetreatment ofvoice in the writings of Latingrammarians, see Hovdhaugen1987. For some descriptionof the commu¯nia, see Draeger 1878and Flobert 1975; they are alsonoted in passing in most handbooks anddescriptive grammars. Finally,McCartney (1926) illustrates several ways inwhich deponent verbs were connected with‘ ‘passivemeaning,’ ’ incases inwhich passive syntax was notan option. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 195 voicesystems in theseterms). 2 1 Thedetails of this relationship will be madeclear as the discussion proceedsand the status of thefeature [pass] isclarified. The point to be stressedis that deponents andverbs in passive syntax are related to the same abstract feature [pass], which results in the realizationof passive forms. Inspelling out this analysis, I willthus be addressing two facts: first,the fact that deponent verbs have only passive form; and, second, the fact that the form of theperfect with passives and deponents is analytic. As thearguments presented in this section show,analytic perfects can be realizedin transitive, active syntax; that is, the analytic /synthetic distinctionis based on thefeature [pass], not on passive syntax per se. The most direct analysis ofthisfact is one in which the differencebetween perfect forms isisolated in asinglefeature; with theexception of thisfeature, the analysismust employ essentially identical syntactic structures for bothanalytic and synthetic forms. Inthe following sections I presenttwo distinct ways in which the feature [pass] canbe treated.The first treats [pass] asa featureof the morphology only, following the idea that only morphologicalfeatures can be arbitrarily associated with Roots. After showingthat this type of solutionis problematic, I presenta pairof treatments, in which [pass] isvisible in the syntax. Beforepresenting these specific analyses, however, I willoutline some assumptions about basic clausestructure in Latin.

4SyntacticStructure ofthe Clause Inanalyzing the structure of theLatin verb, I willassume a clausestructure consisting of three functionalheads, T(ense), Asp(ect), and v, asshown in (12). (12) Syntactic structure (hierarchical; not linear) TP

T AspP

Asp vP

v Ï P

Ï §ROOT§ DP

21 The‘ ‘passive’’ inflectionis notrestricted tosyntactic passives, but also appears inanticausatives (intransitives inthe causative /inchoativealternation) and certain typesof reflexives.This is oftenreferred toas aresidual‘ ‘mediopassive’’ use of the -r form(from Leumann, Hofmann, and Szantyr 1963:sec. 390). (i) induo¯/induor‘ puton’ lavo¯/lavor‘ wash’ veho¯/vehor‘ carry’ verto¯/(re-)vertor‘ turn’ 196 DAVIDEMBICK

The head v hereis thelight verb that figures in a numberof recent syntactic and semantic discussions(see, e.g., Hale and Keyser 1993, Kratzer 1993,Harley 1995, Marantz 1995, 1997, Harleyand Noyer 1998, Chomsky 1995, 1998, Lidz 1998, and McGinnis 1998 for somerelated perspectives).Features here relate to agentivity and causativity, eventivity /stativity,licensing of externalarguments, and certain syntactic features such as Case. Thelicensing of externalarguments is ofparticular relevance in the Latin voice system; in particular,the systematic appearance of passive forms inLatin is correlated with this feature. 2 2 However,this correlation does not work in bothdirections, as shownclearly in the discussion of transitivedeponents above. Thus, the lack of anexternal argument implies the feature [pass] will bepresent,but the converse is not true. The exact nature of the correlation between the external argumentproperty and [pass] dependsupon whether [pass] isonly inserted in the morphology orispresentin the syntactic derivation; this will be clarifiedin thefollowing sections. The point for nowis that the conception of v ascontaining different combinations of features of this type allowsfor (amongother things) a directtreatment of the syntactic conditions correlating with [pass] inLatin. Different feature combinations will be presentunder v, butthe ones that have in commonthe lack of an external argument (i.e., passives and unaccusatives) will be identical morphologically.Passives will have an agentive v, alongwith no external argument, whereas unaccusativeswill have a nonagentive v alongwith no external argument. The common property oflackingan external argument, which correlates with [pass], accounts for theidentity in form. Aspcontains features relating to perfectivity and imperfectivity: the completeness or incom- pletenessof the eventuality associated with the verb and v. Tcontainstemporal features such as [past].Finally, the notation Ï hereis for the Root, thatis, the member of theopen-classvocabulary appearingin this position. The reasons for representingthe Root with this neutral (e.g., category- free) notationwill become clear as the discussion proceeds. 2 3 Theseparate status of the heads T andAsp is mostclearly seen in thepluperfect, as inthe followingform: (13) ama¯veram am- -a¯--v- -er- -am loveTh Asp T Agr ‘Ihadloved . ..’ Theassumption is that in the default case there will be movement up to T; thisresults in thepattern of synthetic verb forms seenin tenses other than the perfect. Following movement of

Theview that the - r forms figurein anumberof systematic alternationsapart frompassivization is also presented inBaldi 1976. 22 There are mostlikely some unaccusativesin Latin that appear inactive form;this is neverthe case withpassives, however.Although the reasons behindthis ‘ ‘splitbehavior’ ’ ofunaccusatives are ofinterest from the perspective of argumentstructure, they cannot be addressedhere. 23 One furtherposition is theTh(eme). Thetheme vowelof the Latin verb is, I assume, theinstantiation of a position addedin the morphology, onto which conjugation class features oftheRoot are copied.In the tree itappears as Th on the Ï -v unit.For a recent approachto theme vowelsin theDistributed Morphology framework, see Oltra 1999. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 197 the Root to v andAsp to T, Iassumethat an Agr(eement) node is added in the morphological component,to produce the object shown in (14). 2 4 (14) The Latin verb T

Asp T

Ï -v-Th Asp T Agr As illustrationof two synthetic forms thatare realized in this structure, consider the active andpassive imperfect of theverb laudo¯ ‘praise’. (15)a. laud-a ¯-ba-t ‘he/shewas praising /praised. ..’ b. laud-a¯-ba¯-tur ‘he/shewas beingpraised’ Thesetwo forms willbe realizedin thestructure shown in (16),as willthe imperfect forms ofdeponentverbs (e.g., hortor, imperf. hort-a¯-ba¯-tur). (16) Structure/Features T

Asp T

Ï -v-Th Asp T Agr

imperf [past] [3sg] (pass) [pass] Thetwo cases will be for themost part identical in featurecontent, differing in whether or notthe feature [pass] ispresent in Asp (see belowfor thelocation of [pass]). Inthis structure theRoot position Ï willbe instantiated by laud-,andthe theme position, occupied by the conjugationclass feature I, bythe theme vowel - a¯-.Onthe assumption that the imperfect aspectual

24 Theanalysis is incompatible with one in which separate projectionsappear foragreement morphemes,as inPollock 1989.If agreement were locatedsyntactically in agreement projections(AgrP), the analytic and synthetic perfects would haveto beassociated withdistinct syntactic structures; the former withtwo AgrPs, the latter withonly one. For discussion ofissues surroundingAgrP, see Iatridou1990 and Chomsky 1995. 198 DAVIDEMBICK featureis realizedas zero, the tense feature, common to both forms, willrelate to the following vocabularyitem: (17) -ba¯- [past] $ TheAgr nodewill then be realizedas follows: (18) -t 3sg. $ Theresult in this case is active laud-a¯-À-ba-t. Therealization of synthetic passive forms ismore complicated. The reason for thisis that whereasthe feature [pass], which is directly related to the passive forms, appearsin the Ï -v domain,or possiblywith Asp, [pass] isrealizedin theT-Agr area,typically in an - r-component. Theimperfect is usedto illustrate here. 2 5 (19) Imperfectpassive laud-a¯-ba-r ‘Ipraise’ laud-a¯-ba¯-ris/(-re) ‘youpraise’ laud-a¯-ba¯-tur‘ he /shepraises’ laud-a¯-ba¯-mur ‘we praise’ laud-a¯-ba¯-min¯õ ‘youpraise’ laud-a¯-ba-ntur‘ theypraise’ Oneoption that is available is to treat the passive endings as indivisible units, which are specifiedcontextually to instantiate agreement features when [pass] ispresenton Asp. (20) Provisionalrealization of passive Agr 1sg. -(o)r / [pass] $ 2sg. -ris / [pass] $ 3sg. -tur / [pass] $ Thereis a weaknessto this approach, however; it fails to account for thefact that the majorityof the passive endings include components that are found in activeAgr aswell (common componentsare boldfaced in the passive Agr). (21) Passiveand active 1sg.passive -( o)r; cf.active - o¯ 2sg.passive -ri s;cf.active - s 3sg.passive - tur;cf. active - t 1pl.passive - mur;cf. active - mus 2pl.passive -min ¯õ ;noactive counterpart 3pl.passive - ntur;cf. active - nt

25 The form -re of2sg. was prevalentin earlier Latin;the form - ris isin fact regardedas theoriginal - re, combined with the -s fromthe active 2sg.(Sihler 1995:475). FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 199

Thedanger is thus that the treatment sketched in (20) misses a generalizationabout components commonto the two types of Agr. Adetailedanalysis of this issue is presented in Embick and Halle 1999; here I willsimply outlinethe basics of ananalysis that segments the passive forms. Theessential component to an alternativeanalysis involves treating - r-asthe basic realization of thefeature [pass]. Thelocation of the feature [pass] dependsupon the analyses of this feature to bepresented insubsequent sections. For now,the point is that if the only starting positions for the[pass] featureare below T-Agr, thestructuremust be readjustedto place [pass] inT-Agr. Onthis scenario, inorder to accountfor theposition of [pass] inthe verb (i.e., adjacent (left or right)to Agr) two mechanicaloperations are required: first, the [pass] featuremust be separated from Aspthrough theprocess of Fission; and,second, it must be adjoined to Agr, viaa form of Morphological Merger (see below). 2 6 Themodified structure, following the relevant operations, is shownin (22). (22) Structure after operations T

Asp T

Ï -v-Th Asp T Agr

[imperf] [past][pass] Agr

[3sg] Thisprovides a positionin which the - r ofthepassive can realize the feature [pass] directly. Inorder to account for thefact that the - r componentfollows Agr in1sg., 1pl., 3sg., and 3pl., whilepreceding it in 2sg., we canappeal to thefact that the structure in (22) is hierarchical,and linearizedas Agr-pass inthe default case, but as pass-Agr when2sg. is present. Afullertreatment of the Latin conjugation would be required in order to determine the adequacyof the analysis of [pass] infinite forms sketchedabove. For thetime being, however, thepoint is that there are mechanical derivations that will account for theposition of [pass].In thefollowing sections I presentthree analyses of adifferentaspect of thefeature [pass], addressing thequestion of whether it is present only in the morphology, or in the syntactic derivation as well.

26 Thepositioning of [pass] inthe T complexis problematic;the clearest argumentsshow that [pass] islowin the structure(i.e., with v orthe Root), yet its morphological realization is withT-Agr. The suggestion here merely highlights theproblem, which may beresolvedmechanically. In partthe issues are related tothestatus of [pass] inthe first place, which,as willbecome clear insection 5, is complicated. 200 DAVIDEMBICK

5TheStatus of[Pass] Thestatus of the feature [pass] inthe grammar isthe central concern of this section and the next.Earlier sections have established two primary points. First, the feature [pass] underliesthe differencebetween analytic and synthetic perfects. Second, this feature is not exclusively associatedwith passive syntax; it is also associated with deponent verbs, for reasonsthat are completely arbitrary. Thediscussion now turns to three possible treatments of [pass],each of whichhas distinct meritsand problems. The firstsolution presented is morphological, claiming that the feature [pass] arisesonly in the postsyntacticmorphology. This solution maintains the idea that arbitrary features ofRootscannot affect the syntax. However, I showit tobe problematicon thebasis of anargument showingthat [pass] istrulyactive in asyntacticoperation. I thenpresent a secondsolution, based onthe idea that [pass] isan uninterpretable syntactic feature. Although this treatment is ableto capturethe behavior of [pass] andthe differences in theLatin perfect, it does so at thecost of a complicationin the syntax. I thereforeconsider a furtheroption, in which the [pass] featureof deponentRoots is visiblefrom theoutset of the syntactic derivation, by virtueof the Root’ s being insertedearly.

5.1Analysis 1: [Pass] as aMorphologicalFeature 5.1.1Basics of a MorphologicalTreatment Thepoint of proposinga morphologicaltreatment of[pass] isthatit would maintain the claim that Roots cannot be inherentlyspecified for features thatare relevant to the syntax. That is, because the formation of an analyticor syntheticform is basedon afeaturethat is sometimes an idiosyncratic property of certainvocabulary items, namely, theinherent specification for [pass] withdeponents, the formation of the analytic and synthetic perfectsoccurs in the morphology. In the syntax the two would be identical, with movement of the Ï -v complexto Asp (hierarchical structure is represented in (23) only). Theidea is that the factors conditioning the difference between perfects are not exclusively syntacticin nature. The information required to determine analytic or synthetic form isnot avail- ablewith deponents until after Vocabulary Insertion and does not correlate directly with active orpassivesyntax. Both have the structure in (23) when coming out of the syntactic derivation; postsyntacticoperations, presented in the next section, account for thedifferences. 5.1.2Analytic /Syntheticin Morphology Inthe creation of the syntheticperfect from thestructure in(23),a postsyntacticprocess is requiredto bring together the T-Agr and Ï -v-Asp complexes. Thisis the process of Merger (Marantz 1984, 1988; also Bobaljik 1994, 1995, Embick and Noyer1999). Unlike syntactic movement, which, although local, can operate over large structures, Mergeris defined to operate under adjacency. Usedin this sense, adjacency can be stated in termsof hierarchical structure, such that a headcan merge with the head of its complement; 2 7

27 Thelowering of T totheverb in English (i.e., affix hopping) is anexample ofMergerof thistype. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 201

(23) Passive (output of the syntax) TP

T AspP

Asp

Asp vP

Ï -v Asp v Ï P

[perf] t t

(24) Merger TP TP

T T

T Asp Þ T Asp

T Agr Asp vP Asp vP

Ï -v Asp v Ï P Asp T v Ï P

t t Ï -v Asp T Agr t t or,adjacency can be definedlinearly, such that Merger can put together linearly adjacent elements. InLatin it is the first notionthat is relevant. The basic idea behind Merger of this type is that thehierarchical relationship between the T-Agr and Ï -v-Asp complexesis collapsed;it operates asshown in (24) for theLatin perfect (nodes in question are boldfaced). 202 DAVIDEMBICK

Theoperation thus has the effect of reducingthe T-Agr complexand the Ï -v-Asp complex, whichit dominated syntactically, into a derivedstructure [[ Ï -v-Asp][T-Agr]]. Noticethat aspects ofthemovements that have taken place are specific to the perfect; namely, in theperfect, syntactic movementcreates Ï -v-Asp, butdoes not movethis further to T. Inother tenses the entire complex willbe movedto T inthe syntax. Therelationship between this process and the hypothesized feature [pass] isthat Merger isprecluded from applyingin the presence of the morphological feature [pass], whatever its provenance. (25)Merger in (24) is blockedby a [pass] morphologicalfeature. Thismakes concrete the fact that synthetic perfects cannot be formedin thepresence of a[pass] feature—itisnotsimply something about passive syntax that defines the appearance of this form. Thereare two cases in whichMerger may be bled: (a) inthe syntactic configurations associated withthe [pass] feature,and (b) inthe presence of deponents, which bear [pass] inherently.In orderto maintain the identity of thefeature, in (e.g.)syntactic passives, the feature [pass] would beinserted in the morphological component (see footnote34). 5.1.3Location of the Pieces Inthe analysis above, the interaction between T andAsp (based onwhether[perf] and[pass] cooccur)that results in theform oftheperfect occurs in averylocal domain.However, in terms of the surface word order of analytic perfects, the participle and auxiliaryneed not appear adjacent to each other in the surface string (participle and auxiliary italicized). (26) Passives a. Ea adhibi-t-a doctr¯õ na est . . . such-FEM add-to-PART-FEM instruction- NOM be-PRES.3SG ‘Suchinstruction has been given . ..’ (Q., Fr.i.1.7;Allen and Greenough 1931:304) b.Ut exuno ¯quondamin duo ¯spopulo¯s d¯õv¯õ-s-a Albana re¯s asout-of one once into two peoples divide- PART-FEM Albannation est . . . be-PRES.3SG ‘As formerlyfrom onepeople the Alban nation was dividedinto two . ..’ (L.,I.28.7) (27) Deponents2 8 a.Lex Terentilia . ..novo¯s adgres-s-a law-NOM Terentilian. ..new- ACC.MASC.PL menace-PART-FEM co¯nsule¯s est. co¯nsul-PL.ACC be-PRES.3SG ‘TheTerentilian law . ..menacedthe new consuls.’ (L.,III.X.5)

28 Theparticiple adgressa here is anorthographic variant of aggressa, from aggredior ‘attack’. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 203

b. . . . quam s¯õ mih¯õ tum essent omne¯s gra¯tula¯-t- ¯õ. ...thanif me- DAT then be-IMPERF.SUBJ .3PL everyonecongratulate- PART-PL ‘...thanif everyonehad then congratulated me.’ (Cic., Plan. 66) Thisis of interest because the process moving the two parts of the perfect is apparently syntactic,in that it moves elements over large structures. At thesame time it must follow the determinationof asyntheticor analyticperfect. The reasoning is asfollows. The assumption we aremaking is thatthe determination of ananalytic or syntheticperfect only comes about in the morphology(i.e., after the syntax). In order for thisto workproperly, T mustimmediately dominate Aspfor theseheads to be combined by MorphologicalMerger, the process that creates synthetic perfects.However, the examples above show that T andAsp can be moved apart from oneanother. If allmovement happens in the syntax, then the movement responsible for theexamples abovecould potentially move Asp away from Tincases in which a syntheticperfect should be created.The result would be an active, analytic perfect for anormalverb. 2 9 (28) *Lauda¯tus (X) (Y) m¯õ litem est. praised-PART (X) (Y) soldier- ACC be-PRES.3SG ‘He (X)(Y) praisedthe soldier.’ Butforms ofthis type do notappear. The conclusion is thusthat the optional movements resulting inexamples like (26)– (27) must take place after the type of perfect has been determined (i.e., afterthe morphology). Oneinterpretation of this fact would be that there is a typeof constituent movement that appliespostmorphologically (i.e., applies to grammatical structures after Vocabulary Insertion hastaken place). The status of sucha positionis difficult to assess in theabstract. In general, the statusof optionalmovements of anytype is somewhat unclear in currentsyntactic frameworks. 3 0 Allowingthis postmorphological movement would result in atheoryin whichthere are effectively twosyntactic components: first, the pure syntax, operating premorphologically; and, second, a form ofsyntax that applies after Vocabulary Insertion and effects stylistic movements. In the absenceof strongarguments, this is problematic. The surface position of elementsmay potentially sayvery little about the syntax, given that their placement could have been brought about in the phonology(i.e., after the syntax and certain morphological operations). Unless the postmorpholog- icalsyntax could be shownto have properties that differ clearlyfrom thoseof thepremorphological syntax,its existence is questionable. 3 1 Thispoint is clearwith reference to theprocess of Merger discussedabove. The theory does allow for thispostsyntactic process to effect movementlike operations.However, Merger is unlike syntactic movement in beingdefined to apply under adja-

29 Thisassumes thatthe morphology does not filter— that is, that it interprets syntactic structures only and does not rendersome syntacticstructures ungrammatical. 30 Forinstance, Chomsky (1998:21) hypothesizes that stylistic movements might occur late onthePF branch. 31 Itshould be noted,however, that PF movementsperforming syntaxlike operations have been proposed in other contexts;see, forinstance, Sauerland, to appear, and Elbourne 1999. 204 DAVIDEMBICK cency,under limited conditions— primarily repairing morphological or phonological problems (e.g.,satisfying clitic or affix dependencies). It is thus clearly not a secondsyntax in the sense thatthe process required to generate examples like (26)– (27) would be. Theposition I willtake is that the movements found in (26)–(27), which look like syntactic movement,are in factthe result of asinglesyntactic computation, which occurs prior to morphology.That is, there is no postmorphologicalsecond syntax. Part of thejustification for thisstance istheidea that the existence of aPFmovementof thetype discussed above must be proven,not assumed.3 2 Theconsequence of thisposition for thestatus of [pass] isclear: because the existence ofsuchmovement follows from themorphological treatment of [pass] developedin thissection, we havean argument that this solution should be abandoned in favor of a syntacticstatus for [pass].

5.2Analysis 2: [Pass] in the Syntax Thefollowing points must play a rolein asyntactictreatment of thefeature [pass]. First, as noted already,the feature [pass], which is responsible for passivemorphology and the form ofthe perfect,does not directly bring about passive syntax whenever it ispresent;the argument based ondeponents makes this clear. Furthermore, there is no basis for interpretingthe feature [pass] inallof itsoccurrences, as this would require deponents and true syntactic passives to havesome commoninterpretive quality, contrary to what was demonstratedabove. Giventhese criteria, one possibility is totreat [pass] asan uninterpretable syntactic feature andto make the syntactic difference between the perfects result from this.In outline, the analysis of[pass] asan uninterpretable feature in the syntax captures the behavior of the perfect by preventingmovement of [perf] Aspto T. 3 3 (29)[Perf]Asp doesnot move to T when[pass] ispresent. Thestructures in (30) and (31) are then relevant for thetwo perfects. Incases in which the syntax is ‘‘passive’’ (i.e.,in which there is no externalargument), the appearanceof thefeature [pass] canbe tiedto the features of v. Thatis, just as thereis a v related toactive, transitive syntax, and with a particularfeature content, so there is a v associatedwith structuresin whichonly internal arguments are licensed. The idea is then that the feature [pass] ispresentbelow v when v doesnot license an external argument syntactically. 3 4 Otherproperties

32 Thus,if it becomes clear thatsuch movements are necessary, themorphological treatment of[pass] outlinedin section5.1.2 will become anoption. In order to prevent the derivation of analytic perfects forregular verbs in active syntax,however, something additional is required. 33 Theaccount of Giorgiand Pianesi (1991, 1997) takes theanalytic /syntheticdifference tofollowfrom the categorial statusof theirprojection T2 (basically Asp here) incombination with properties of passivesyntax. As shownin section 3,this treatment is unworkable;analytic or synthetic form has nothingto do with passive syntax per se. Ina sense, however,Giorgi and Pianesi seek toderive the surface categoryfrom the position to whichmovement has occurredin thetree; tothis extent, their account is similar tothat presented here. 34 As opposedto systems inwhich voice features are assignedpostsyntactically, in a syntacticconfiguration with noexternalargument; see Embick1997, 1998 for discussion of suchoperations, and for discussion of the syntactic and semantic features of v foundwith passives. For recent perspectiveson thesyntax associated withthis type of v, see also Lidz1998 and McGinnis 1998. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 205

(30) Synthetic (31) Analytic TP TP

T T

T AspP T AspP

Asp T Asp Asp

Ï -v Asp Asp vP Asp vP

[perf] t v Ï P Ï -v Asp v Ï P

t Ï (DP) [perf] t Ï (DP)

t t of v, relatingto agentivity for instance,will determine whether the resulting structure is passive orunaccusative. Withdeponent verbs, the feature [pass] isrequired for theinsertion of the relevant Roots. Thiscan be captured by assuming that a [pass] featureis generatedsyntactically in the position ofthehead of the complement of v (i.e.,the position into which Roots are inserted). Effectively, thedeponents are licensed for insertiononly into Root nodes containing the feature [pass] (see Harleyand Noyer 1998 for discussionof the licensing of Roots).

(32) DEPONENT Ï [pass] $ Becauseof thiscondition, deponents will not be insertedinto ‘ ‘normal’’ Rootpositions; they will appearonly when [pass] hasbeen generated on that position. In addition, the feature [pass] below Ï isnotassociated with any syntactic effect; the v abovemay be transitive and active, as it in factis withtransitive deponents. Inaddition to the above points, there is thequestion of the location of thefeature that the Ï -v [pass] featureswill be checkedagainst. That is, the uninterpretable [pass] featurein the Ï -v domainmust be checked by a highercategory that attracts [pass] inorder for thederivation to belegitimate. For concreteness,I willassume that it is T thatattracts [pass] intensed clauses; otherwise,Asp (nontensed contexts). Although not a greatdeal hinges on thischoice, the realizationof passivemorphological forms inLatin is associatedwith the T-Agr position,suggesting a 206 DAVIDEMBICK connectionbetween T and[pass] thatcan be capturedby specifyinga featuralrelationship between the two. Theeffect that the [pass] featureshave on syntax is simple. When [pass] and[perf] areboth present,movement to T from Aspis impossible. Thus, Ï -v remainin Asp. As aresult,there is ananalyticform inthe perfect whenever [pass] ispresent. The actual derivation of forms inthis structurewill be undertakenin the nextsection. For now,there is the question of howcombinations of[pass] featuresin the syntax are checked. Given the proposals above, there are three positions thatfigure in derivations with [pass]: (33)a. T (attracts /doesnot attract [pass]) b. Under v, associatedwith the lack of anexternal argument c. Under Ï ,thepositionof theRoot (i.e., the position where the insertion of deponents islicensed) Thetwo types of Therehave to dowith the realization of passiveforms onT-Agr; thatis, theT thatattracts [pass] resultsin whatis effectivelypassive agreement. The overt manifestation ofthis is in the realization of the - r-form passiveagreement suffixes noted earlier. Within the verb,these appear in the T-Agr complex;and this follows naturally if Tattracts[pass] incertain cases.The ‘ ‘agreement’’-type features relating to [pass] onTarethen expressed on T-Agr. Syntac- tically,this type of Thasto be accompanied by a [pass] featurein the Ï -v domain.3 5 Itis important to emphasize that only the [pass] generatedon v isdirectly related to the syntaxof passivization. The other sites in which the [pass] featurecan be generated are not specificallycorrelated with the lack of an external argument. Interms of theirinteractions (i.e., in terms of whathappens in derivations in which [pass] appearsin different combinations of thesepositions), there are the cases in (34) to consider. 3 6 (34) Derivationswith [pass] Name Tproperty v property Rootproperty Case1 Tdoesnot attract [pass] no[pass] under v no[pass] under Ï Case2 Tdoesnot attract [pass] no[pass] under v [pass] under Ï Case3 Tdoesnot attract [pass] [pass] under v no[pass] under Ï Case4 Tdoesnot attract [pass] [pass] under v [pass] under Ï Case5 Tattracts[pass] no[pass] under v no[pass] under Ï Case6 Tattracts[pass] no[pass] under v [pass] under Ï Case7 Tattracts[pass] [pass] under v no[pass] under Ï Case8 Tattracts[pass] [pass] under v [pass] under Ï

35 Whenmovement to T is blockedin the [pass] perfect, thisrequirement of T wouldbe met byvirtue of the relationshipAgree, inthe sense ofChomsky1998, that is, without movement of the [pass] feature toT inthe syntax. Allother things being equal, this would predict that passive forms appear with esse, contraryto what is actuallyfound. Thus,something additional is needed here ifthisapproach to [pass] is tobe maintained. 36 The verba commu¯nia, theverbs noted above as havingonly passive form but either passive or active syntax, requirean additional assumption. In these cases there wouldpresumably be a[pass] feature ontheRoot position, as well as one on v incases ofpassive syntax. Mechanically, the simplest treatment inthis case wouldbe one in whichthe two FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 207

Theresults of each of these derivations, along with the verb types they correspond to, are as follows: Case 1: Normalsyntax, in which deponent verbs will not appear. Case 2: Deponentverb licensed in Root position; the uninterpretable feature goes unchecked, andthe derivation crashes. Prevents deponent verbs from appearingin syntax without T[pass]. Case 3: Syntaxwith no externalargument, but in whichthe [pass] featureis uncheckedand thederivation crashes. Case 4: Passivesyntax in which only deponent verbs are inserted; the [pass] featureis not checked,on the assumption that [pass] on v and[pass] ontheRoot do notinteract; see case 8. Case 5: Passiveforms withneither passive syntax nor a deponentverb. The requirement of Tisnotmet, and the derivation crashes. Case 6: Thederivation for deponentverbs; these are licensed in the Root position, and the [pass] featureis checkedby T. Case 7: Passivesyntax with normal verbs; the [pass] featureon v ischeckedagainst T. Case 8: Assumingthat the [pass] featureson v and Ï donotinteract with one another, this isthe case of verbswith only passive forms appearingin passivesyntax (the verba commu¯nia discussedearlier). Mechanically,this solution provides a meansof treating [pass] asa syntacticfeature, while maintainingthe idea that it is associated with certain Roots inherently. In addition, by treating [pass] asan uninterpretable syntactic feature, the solution maintains Feature Disjointness. This treatmentincurs a cost,however. The syntax is complicated by theworkings of thefeature [pass] in (34).3 7 Andthe syntactic effect that motivated a syntactictreatment of the feature [pass] inthe

[pass] features of Ï -v are effectivelytreated as oneforthe purposes of checking with the [pass] onAsp.Something like thismust be stipulated in any case, toprevent a derivationin which a [pass] feature onthe Root position licenses a deponentand is checkedagainst a [pass] on v. Theaccount presented here has three possiblepositions for [pass]; restrictingthe positions in which the feature may begeneratedhas otherconsequences. For instance, one alternative would involve restricting the lower [pass] feature to v. Thiswould simplify part of thesystem, inthat there wouldonly be one lower locus for the [pass] feature; deponents mightthen be specifiedas follows: (i) DEPONENT Ï / [pass] $ However,this treatment raises seriousquestions about the nature of competitionfor Roots. All other things being equal,the specification in (36) would predict that deponents should always winout over normal verbs for insertion when [pass] is present.This predicts that normal verbs would have no passiveforms, which is clearly incorrect.The possibility existsthat competition for Roots simply behaves differently from the competition found with functional morphemes. This positionhas clear ramificationsfor other aspects ofthe theory, in the treatment ofsuppletion,for instance (see section 5.3).But the scope of the issues raised bythisstance take itbeyond the concerns of thisarticle. Anotherproblem for this type of treatment concernsthe status of [pass] on v. Inthe analysis presented in the preceding section,[pass] on v is always correlated witha syntacticeffect— the failure to license anexternal argument. On the treatment justoutlined, this is nolonger the case. Amongother things, the question of how to prevent normal verbs from beinginserted into ‘ ‘deponentsyntax’ ’ isunresolved. 37 Thetreatment of[pass] as anuninterpretable feature alsomakes certain predictions,the status of which is not clear. Forinstance, it seems topredict that deponent Roots will not appear in‘ ‘simple’’ nominalizations,that is, nominaliz- ationswithout an Asp head. The reason for this would be that the [pass] feature requiredfor the insertion of thedeponent wouldnot be checked, if there were nohigher head. 208 DAVIDEMBICK firstplace— namely, the nonmovement of [pass][perf] Aspto T— iscapturedorthogonally. That is,the restriction to the effect that movement of [pass][perf] Asp toT doesnot occur requires onlythat [pass] bevisiblein the syntax. The further claim, that [pass] isanuninterpretablefeature, arisesonly as a resultof treating Roots as subject to LateInsertion; because features of the Root itselfcannot be visible, an uninterpretable feature is the only remaining candidate to associate withthe effect to be captured.With the hypothesis that Roots are not visible in the syntax, this was theonly option for making[pass] syntacticallyactive. In the following section I presenta secondsyntactic option: the treatment of [pass] asa featurevisible in the syntax, with specific Rootsvisible in the syntax as well.

5.3Analysis 3: A SyntacticTreatment with Roots in the Syntax Thetreatment offered above is based on theargumentthat [pass] hassyntactic effects. As noted,the analysisaccording to which [pass] isanuninterpretablefeature is only one possibility;moreover, it hasshortcomings, in the sense that it complicates the syntax somewhat. A secondidea, one that avoidsthe complications of checking this feature, would be to hold that the feature [pass] is actuallyinterpreted when it appears on v, butnot interpreted when it appears arbitrarily with deponents.The result of thisinterpretation would unify passives and unaccusatives, whose com- moncomponent is the lack of an external argument. 3 8 Onthis approach, deponent Roots bring thefeature [pass] intothe syntactic derivation, by virtue of being visible in the syntax; that is, underthis option (at least deponent) Roots are inserted early. 3 9 (35) Representationof a deponent (provisional) Ï ROOT[pass] Recallthat [pass] hasa dualnature; it is requiredwhen no external argument is present, but theconverse is nottrue. Transitive deponents appear with external arguments. This behavior can beaccountedfor asfollows. When [pass] isgenerated on v, itaffects Merge: no externalargument ismergedwith v when[pass] ison v. However,the complement of v—namely,the Root and its argument—willcontain [pass] ifthe Root is deponent. From theperspective of v, the [pass] featurein its complement has no effect on its syntactic possibilities; a v canhave an external argumenteven when [pass] appearswith the Root. Thus, there is nocontradiction with transitive deponents.Subsequently in the derivation, however, the Ï -v complexwill contain the feature

38 Of course,this regards this property uniting passives andunaccusatives as havingreal semantic content;and this is contentious.The point is that by treating the feature inthis way, there is noneed for checking, in contrast with the previoussyntactic analysis. Ultimately the status of this assumption will depend upon an understanding of the nature of the‘ ‘noexternal argument’ ’ propertythat [pass] encodes. 39 Therealization of the functional vocabulary would still be postsyntactic; in certain keyrespects theresulting picturewould be similar tothe model presented in Halle 1990.Furthermore, the argument here is onlythat the features of certain (notall) Rootsneed to be visible in the syntax. Itmight also be possible to treat theinsertion at the phase level(Chomsky 1998), with Vocabulary Insertion taking place onthese structuresderived in a cyclic computation.Whether or not this would make predictionsdifferent from a theoryin which Roots are presentin the syntax from the outset is unclear. In either case thepoint is that the Root bears asyntacticallyrelevant feature. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 209

[pass],following movement of theRoot to v. From theperspective of subsequent syntactic operations,[pass] inthe Ï -v domainwill have an identical effect, whether it originates on v or with theRoot. Specifically, movement of Asp to T willbe blocked in each case. This is effectively allthat must be saidabout [pass], at least as far asthe syntax is concerned.No additional mecha- nismsor assumptionsare required to account for thechecking of [pass]. Iwillcompare the two syntactic treatments of [pass] below.First, however, I turnto afurther positiveconsequence common to both syntactic treatments of thisfeature, which suggests that a syntacticsolution is justified.

5.4A Consequence:Active Perfect Participles Asyntactictreatment of [pass] allowsfor anasymmetry in the Latin participial system, manifested inenvironments without T, tobe accounted for directly.The basic fact is that whereas normal verbssimply lack active perfect participles altogether, deponent verbs do haveperfect participles thatappear in active syntax. The relevant environment is called the ablativeabsolute, in which theparticiple appears with ablative case. The following example, taken from Brugmann1895: 137,illustrates this with the deponent verb polliceor ‘promise’:4 0 (36). ..Sulla¯ omnia pollicito¯ ...... Sulla-ABL everything- NEUT.ACC promise-PART.ABL.SG.MASC ‘Sullahaving promised everything . ..’ (Sall., Jug. CIII,7) Syntactically,Latin simply does not allowthe combination of activesyntax and perfect aspect inparticiples. The restriction here is to participles, and not simply nonfinite forms; infinitives, for instance,allow perfect active syntax (cf. amo¯ ‘love’; ama¯visse/ama¯sse ‘tohave loved’ ). Whatever thereason for thiseffect in participles, the statement of the restriction is something like the following,which is simplya propertyof Latin syntax. 4 1 (37)Asp [perf] notselected by T selects[pass]. Thatis, it requires[pass] initscomplement. Recall now that [pass] canbe presentfor oneof two reasons:it canbe either assigned syntactically or associatedinherently with deponent verbs. Onthe first syntactic solution, with regular verbs in active participial syntax, if [pass] is presentwith [perf], therewill be no corresponding [pass] featureon v. Thus,the feature [pass] onAsp will be unchecked,and the derivation will crash. With the special Root-licensing [pass]

40 Example(36) shows an object bearing the case normallyassigned by theparticiple, here accusative. Theactive pastparticiple appears inthis type of syntax(with the normal object case) onlyin Sallust and after (see Brugmann1895). Theimportance of active perfect participlesof thistype was pointedout by AndrewCarstairs-McCarthy in a communi- cationto Rolf Noyer, and I gratefullyacknowledge this lead. 41 Thestatement ofthisrestriction operates ontheassumption that T is notpresent in such structures. The restriction itself is notmeant toexpress afilterof some type,or anythingbeyond a statement ofthefacts, presumablyreducible to somethingelse. Exploringthe nature of restrictions of this type is, of course, an interesting objective in its own right. Butthis will not be undertaken here. 210 DAVIDEMBICK feature on Ï (i.e.,below v),thiswill not happen. This feature will check the [pass] featureon Asp,and this will be acceptablesyntactically, even though the syntax is active.Of course,only deponentverbs will be insertedinto Ï [pass].It thus follows that only deponent verbs will have perfectparticiples in active syntax. 4 2 Onthe second treatment, (37) will simply be violatedif the syntaxis not passive or ifa deponentverb is notpresent.

5.5Syntax, Roots, and Features Themorphological treatment of [pass] presentedin section5.1 is ableto capture the behavior of thefeature [pass] withoutsyntactic complication, and it maintains Feature Disjointness; but it requiresa theoryin which syntaxlike movement takes place in the morphology or after. The treatmentof [pass] asan uninterpretable syntactic feature allows for theforms ofthe perfect and thebehavior of deponentsto be capturedmechanically; it alsomaintains Late Insertion of Roots andFeature Disjointness. Yet it does so at the expense of complicating the syntax, with the additionalpoint that the complication is forced by the assumption that Roots are not present in thesyntax. I thereforepresented an alternative solution to the problem, in which (at least the deponent)Roots are visible in the syntax. This treatment is merely one implementation of the ‘‘EarlyInsertion’ ’ approach.The point to bestressed is that considerations of FeatureDisjointness arewaived under treatments of thistype. 4 3 Thearbitrary features of Rootscan have a cleareffect onthe subsequent syntactic derivation. If thisis the case, then the question that must be asked isalong the following lines: if Root-specific features like [pass] canplay a rolein the syntax, thenis the same true of purely phonological properties of Roots?Or semanticfeatures? There is nothingin the architecture of the theory to preclude these possibilities. Whatis at issue is the status of what is represented in the syntax with Roots. It could be thecase, for instance,that the Root is represented simply with a label—thatis, as Ï 369, where thislabel is realizeduniquely with phonological features, postsyntactically. In this representation phonologicalproperties of theRoot would not be visiblein thesyntax; but other features, which mightaffect the syntax, would be. A treatmentof Rootsalong these lines is accountable to other factors.For instance,the implementation of aparticulartheory of suppletion,presented in Marantz 1995,relies on theidea that Roots are subject to LateInsertion. Marantz holds that suppletion is impossiblefor trueRoots; the reasoning is thatthe entry for asuppletiveRoot, with a contextual conditionon its insertion, would necessarily block all ‘ ‘normal’’ Rootsfrom appearingin the

42 Thepresent participle behaves differently: Latin has nopassive present participles at all,although it does have active presentparticiples (e.g., from laudo¯ ‘praise’, lauda-ns ‘praising’; withdeponents, e.g., hortor,horta-ns ). The restrictionhere is againsyntactic: no passive syntax in present participial syntax. Notice thatif deponents like morior ‘die’are treated as unaccusative,then unaccusative syntax is allowedin the present participle structure: moriens ‘dying’. Of course,how restrictions of this type may bereducedto other syntactic principles is anopen question. 43 At least inpart. Recall thatFeature Disjointnesshas twodirections: (a) arbitrarymorphological /phonological propertiesdo not affect thesyntax; and (b) syntacticosemantic features are notreinserted in the morphology. The first isbeing questioned here. It is worth noting that the second direction makes distinctand interesting predictions about the interface as well. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 211 relevantenvironment. 4 4 If thistheory, which proceeds from thepositionthat there is no suppletion for Roots,is correct, then the considerations presented above will have an effectupon its implementation. Clearlythe questions surrounding the Late Insertion of Rootsare quite involved and require detailedinvestigations in anumberof domains.But the discussion of thissection clarifies several oftheissues that are at stake.The reason that these issues are so clearly visible stems from the startingpoint of theinvestigation: Feature Disjointness, a strongposition on the nature of features andthe interfaces of syntax.As aresultof operating in terms of this hypothesis, we nowhave asituationin which the boundary conditions for furtherattempts to refine the architecture are providedby both syntactic and morphological considerations. Ultimately, many of these issues areempirical; and they will only be resolvedwith the careful analysis of additionalcases, assessed inconjunction with the case presented here. Recalling that the interest of theLatin case emerges asa resultof proceedingwith the most restrictive hypothesis concerning Roots and syntax, namely, LateInsertion and its correlate Feature Disjointness, the path is relativelyclear. Specifically, even ifone is willingto countenanceEarly Insertion in this case, one must proceed with the assumption thatfeatures of Rootsare not syntactically active, absent a demonstrationto the contrary. Under thisstance, the Root properties that are absolutely required to be visible in the syntax will be madeclear; and this will allow for amoredetailed theory of howRoots, syntax, and morphology relateto one another. Restrictingattention now to the Latin perfect, there is a factorthat unites the treatments givenfor Latin;on eachapproach the point that distinguishes the analytic and synthetic perfects isultimately structural and concerns movement to Tinone case, but not in theother. I therefore turnnow to therealization of the forms ofthe perfect and the implications that this analysis has for notionsof category.

6 Category Theanalysis of the analytic /syntheticsplit above relies on the idea that the appearance of a participlein the analytic perfect is not the result of a participle-specificoperation. Rather, the twoare realized in the same basic structure, with movement to Tinone case (synthetic), but not inthe other (analytic). The claim is thatthe formationof the participle then follows from principles ofLatinmorphology, specifically, those concerning the spell-out of functionalcategories contain- ingsyntacticosemantic features. In this section and the next I providean explicit treatment of theseprinciples. 4 5

44 Thus,for instance, if the plural form of dog were hound-s, thefollowing entry would be required: (i) Ï HOUND / [pl] $ Thisis more specific thanthe entry for normal, nonsuppletive Roots and would thus preclude their insertion into the [pl] environment;it would thus be impossible to realize theRoot (e.g.) Ï CAT inthe context of [pl],because theentry in (i) wouldalways winthe competition. 45 Before proceedingwith this, I willpoint out that this treatment, as well as thearguments presented above, follow forthe Latin perfect. Thestatus of the perfect crosslinguisticallyis anentirely different matter. Inlanguages such as English,in whichthe perfect is always analytic,arguments of the type presented above simply cannot be formed. 212 DAVIDEMBICK

6.1Categories and Roots Theanalysis I havepresented provides evidence for asyntactictheory in whichsimple category labelslike V andA areirrelevant. Recall that the information required to determine which type ofperfect is to be formed is localized in the feature [pass]. Based on the effects of [pass], the twotypes of perfectdiffer only in the position to which the Ï -v complexhas raised in the tree: toAsp in one case (31), beyond Asp to T inthe other (30). Otherwise the two structures are identical,containing the functional heads T-Asp- v. Thefact that distinct categories are found in thetwo perfects, auxiliary ` participle(A) inone case, V inthe other, shows that identical clausalstructures may lead to distinctsurface categories. 4 6 Thatis, there is nodifferencein labels ofthetype V for verb,A for adjective( 4 participle),or somedecomposition resulting in these. Whatthis means is that labels of this type are merely convenient tags for morphosyntacticnotions: V 4 rootcombined with tense, N 4 Rootcombined with case /number/gender, A 4 Root combinedwith case /number/gender,with the number and gender (and case) coming from elsewhere. 4 7 Whateverutility these labels might have as shorthand,the point is thatthe Roots bear no simple substantivefeatures on asyntacticosemanticlevel corresponding to thesemorphological notions ofcategory. The differences in the objects listed above, the different ‘ ‘partsof speech,’ ’ must thereforebe reducedto other factors. Theargument that category is a notionrelevant only in themorphology converges with the ‘‘category-free’’ treatmentof nominalizationsfound in Marantz 1995, 1997. Marantz argues that underlyingparts of speech like verb (e.g., destroy), noun (destruction ),andso on, is a Root 4 8 (Ï DESTROY),whichis unspecified for syntacticcategory. Inthis implementation part-of-speech labelsare relevant only in the morphological component. That is, a nounis a Rootin a local relationshipwith a particularfunctional head, D. Thefunctional heads are identifiable in terms oftheir syntacticosemantic feature content and thus play a definingrole in the morphological realizationof Roots.In what follows I willimplement an analysis in termsof Rootsof thistype; thegoal is to reduce the differences associated with part-of-speech labels like N, V, andA to differencesin structuralenvironments and morphological processes applying to Roots.That is, I willderive the effects of differencesin morphological category without assuming a simplesyntac- ticfeatural difference of the 5N/5Vtypein the first place.

6.2The Synthetic Perfect Recallthat when [pass] isnot present in the perfect, the Asp complex incorporates into T, asit doesin other tenses. This is illustratedin (38).

46 Thisis notmeant toimply that participles are always realized instructures identical to those found with finite verbs;rather, they may sometimes appear insuchstructures, as well as instructures in which finite verbs do notappear. 47 Interms ofLatinmorphology, the statement ofthese classes takes thefollowing form. Agr nodes that are attached to[finite] T agree interms ofperson /number.All other elements bearingan Agr-type desinence, whether classified as nouns,adjectives, or participles,will show gender /number/case suffixes,further conditioned by declensional class features. Thisis completelygeneral, as adjectivalendings are asubsetof thenominal endings in Latin. 48 Forthe background of thistype of approachto nominalizations, see Chomsky1970. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 213

(38) Synthetic TP

T AspP

Asp T Asp

Ï -v Asp T Agr Asp vP

[perf] t v Ï P

t t Therealization of the finite perfect in this structure involves Asp with the feature [perf] beingspelled out in a numberof ways,as detailed above (see footnote 7). For instance,with the relativelyfrequent - v- and -s-suffixesfor theperfect, the following spell-out rules are relevant: (39) -s- [perf] / T, List $ -v- [perf] / T $ Otherconditions specifying the other allomorphs of theperfect can be stated similarly. The pointis that these spell-out rules are specifically tailored to the circumstance in whichAsp is in acomplexmerged with T-Agr. Thiswill separate these cases from casesin which Asp is not combinedwith T, whichare realized as the analytic perfect. 4 9

6.3Realization of the Analytic Perfect: Initial Points Thequestions to be addressed now center on the analytic perfect; recall that this is realized in thestructure shown in (40).

49 Previousanalyses, as well as some traditionaldiscussions, have argued that the synthetic forms ofthe perfect havea suffixedform of esse ‘be’in them (see, e.g.,Lindsay 1915, Giorgi and Pianesi 1991, 1997). If thiswere correct, itwould be compatiblewith the treatment Ihavegiven, in the sense thatit would point to a directrelationship between thetwo types of perfect. However,I believethat the evidence actually shows that esse isnotsuffixed in the synthetic perfect forms. Theargument in favor of asuffixed-‘be’ treatment is basedon additional forms oftheperfect: thepluperfect and futureperfect. Iwillfocus on the pluperfect here, as thefuture perfect isequivocal from the perspective of the two approachesto be considered. The pluperfect forms of amo¯, alongwith the imperfect forms of sum ‘tobe’ , are as follows: 214 DAVIDEMBICK

(40) Perfect without movement of Ï -v-Asp to T TP

T Asp

T Agr Asp vP

Ï -v Asp v Ï P

t t

(i) Pluperfectactive (ii) Imperfect of sum 1sg.ama ¯-v-er-am ‘Ihadloved’ 1sg.er-am ‘Iwas’ 2sg.ama ¯-v-er-a¯s‘youhad loved’ 2sg.er-a ¯s‘youwere’ 3sg.ama ¯-v-er-at ‘he /she hadloved’ 3sg.er-at ‘he /she was’ 1pl.ama ¯-v-er-a¯mus‘ we hadloved’ 1pl.er-a ¯mus ‘we were’ 2pl. ama¯-v-er-a¯tis‘ youhad loved’ 2pl. er-a¯tis‘ youwere’ 3pl. ama¯-v-er-ant‘ theyhad loved’ 3pl.er-ant ‘theywere’ Thepluperfect forms thusappear tohavea pasttense formof ‘be’( eram, etc.) suffixedto the stem `-v-(cf. (49)). Likethe perfect, thepluperfect passive involves the - t-/-s-participlewith the imperfect of‘ be’: ama¯-t-user-am. In this way,the pluperfect active appears tobe a(phonologicallyslightly different) parallel towhat happens in the perfect and pluperfectpassive— that is, it appears toincludean overt form of ‘ be’that happens to be collapsedwith the stem ina way notfound in the passive. There are twopoints to be made concerningthis suffixed-‘ be’ analysis. The first is diachronic:it is notthe case that theelement - er-appearingon the perfect stem inthe pluperfect is derivedfrom the verb ‘ be’. Theforms of‘ be’that show -er-are derivedfrom the stem es-viathe process ofrhotacism( s. z. r).Butthe relevant component of the pluperfect can beshownto be derivednot from es-,butfrom a component- is-(fordetails see Sommer 1914,Leumann, Hofmann, andSzantyr 1963). The surface appearance issimilar tothat of ‘ be’owing to the operation of similar phonological processes (inthe case of- is-. -er-,rhotacismand the additional lowering of thevowel). For a discussionof the history of -is-,see Leumann,Hofmann, and Szantyr 1963:sec. 444; for rhotacism, sec. 180;for the vowel change, sec. 86.III. Thesecond point is that there are synchronicdifferences between theelement appearingafter theperfect stem in thepluperfect and the verb ‘ tobe’. Inthe imperfect subjunctivethe form of ‘ be’is basedon thestem es-: essem 4 1sg. imperfect subjunctive.The 1sg. form of the pluperfect subjunctive, on theother hand, is ama¯vissem (i.e.,with - is- and not -es-).No phonological process can beheld accountable here forthe difference, as there is nobasis forraising /e/ before /ss/.So,the pluperfect subjunctive, if itactually included a formof ‘be’, wouldbe * ama¯vessem; butit is not.I thereforeconclude that the pluperfect active islike the perfect active,with an additional element - er-/-is-.Thus,despite surface appearances, thepluperfect active doesnot necessarily givespecial insightinto the structure of theperfect as a whole.Giorgi and Pianesi (1997) propose an alternative to the - is-accountof the pluperfect, acknowledging that their incorporationsolution is controversial. I willretain the more conservativetreatment here,barring any further developments insupport of theincorporation hypothesis. Inany case, theimportant point is thatthe account offered here differsfrom Giorgi and Pianesi’ s approachcrucially interms ofthepoint at whichanalytic and synthetic perfects diverge:based on passivesyntax for Giorgi and Pianesi’ s treatment, butbasedon themorphological feature [pass] here (recall section3). As longas thisdifference isclear, questions concerningthe presence orabsence ofsomething related to‘ be’in the synthetic perfect are ofsecondaryinterest. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 215

Thisbreaks into two components. The first has to dowith the T-Agr complex,and the second with the Ï -v-Asp complex.Beginning with T-Agr, theappearance of esse ‘be’can be attributed toa sortof defaultinstantiation of tense.However, this is different from do-support,in thatit is notdirectly triggered by agreement /finiteness.That is, whereas there is no do-supportin infinitives, wherethere is neverovert tense /agreementto besupported, Latin ‘ be’appears in infinitival forms oftheanalytic perfect: lauda¯tusesse ‘tobe praised’ . 5 0 Themorphology found in the participleis more complicated, as it does not directlyinstantiate voicefeatures like [pass] oraspectual features like [perf]. Rather,it is thedefault realization of thehead Asp. To see this, an excursus into Latin deverbal formations is necessary. TheLatin participle that appears in the perfect, the so-calledperfect passive participle, shows unpredictableallomorphy between - t- and -s-.Thus,consider laud-a¯-t-us ‘praised’(infinitive laud-a¯-re) and ius-s-us ‘judge’(infinitive iub-e¯-re); here -t-/-s-arethe ‘ ‘participial’’ affixes,and -us isthe gender /case/numberdesinence. I willrefer tothe former pairof exponents, as found inthe perfect and forms tobe discussed below, as - t-/-s-throughoutthe remainder of thediscussion. Themain complications arising with - t-/-s-stemfrom thefact that they have a defaultstatus in thederivation of deverbalformations (in this I followAronoff 1994; see below). A long-standing problemin theanalysis of Latinparticiples is the fact that the future active participle (e.g., lauda¯- t-u¯r-us ‘aboutto praise’ ) shows,in addition to the - u¯r-component,allomorphy identical to that foundin theperfect participle (as wellas inother forms, suchas the supine, and various nominalizations).5 1 At thesame time the two differ in voice. The problem comes from anattempt to treat thepast and future participles in auniformway. The forms ofthetwo participles for amo¯ are as follows: (41)Past participle: am-a ¯-t-us‘ loved’ Futureactive participle: am-a ¯-t-u¯r-us ‘aboutto love’ Allomorphyin thesetwo participles, between - t- and -s-,isboth identical and phonologically nonpredictable,throughout a numberof distinct deverbal formations. The forms with- s- were originallymerely variants of the - t-suffixconditioned by regular morphophonological factors (the -s-appearedoriginally with stems ending in dentals).However, the - s-suffixspread to forms inwhich it could not have been brought about phonologically from - t-,makingit a trueallo-

50 Englishdoes, however, require have withthe infinitival perfect. Itshould be noted that the discussion here is meant toapply exclusively to the appearance of esse inthe Latin perfect. Verbal supportof thistype might be reducedto the requirement that T becombined with a v element, realized as the‘ ‘default v’’ esse. Insuch a case thesupport of T providesa furtherargument in favor of the syntactic solution developedabove. If Tsupportwere morphological,then the morphology would be forcedto introduce a default v on T, tobe realized as esse. But v isa syntacticosemanticobject. Introducing it in the morphology would require a system in whichmorphology has theability to introduce objects relevant only in syntax /semantics, anundesirable consequence giventhe idea thatFeature Disjointnessshould be maintained to the fullest possible extent; that is, having the morphology insert v wouldbe a violationof the second direction of Feature Disjointness. 51 See Benveniste1948 for the latter types.As noted,the problem is not restricted tothe participles; the same allomorphynoted below occurs in other types of adjectives andnominalizations. I willconcentrate here onthe ‘ ‘past passive’’ and‘ ‘futureactive’ ’ participles,as thepoint may beseen clearly inthis contrast. The theoretical questions raised bythe behavior of the ‘ ‘pastpassive’ ’ and‘ ‘futureactive’ ’ participleshave been discussed at least since Matthews 1972. 216 DAVIDEMBICK morph.5 2 Thus,the realization of the - s- or -t-componentwill be thesame in the perfect and the futureactive, even with verbs that have irregular participial forms. 5 3 Thus(cf. Aronoff1994): (42) Verb Perfectpart. Future active part. Translation veherevec-t-us vec-t-u¯r-us ‘carry’ iube¯re ius-s-us ius-s-u¯r-us ‘order’ premerepre-ss-us pre-ss-u¯r-us ‘press’ ferre la¯-t-us la¯-t-u¯r-us ‘bear’ Theproblem is thus as follows. In one type of analysis of the future perfect, an affix - u¯r- isadded to the perfect participle. However, the two participles differ in voice (in the normal case):the past participle is passive; the future is active.As noted,this is part of amoregeneral problem:the question of - t-/-s-allomorphyarises in a numberof deverbal formations. Aronoff (1994)argues convincingly that there is no coherentmorphosyntactic content, in terms of voice oraspect,to be associatedwith - t-/-s-. Thisbasic position is one that I willadopt here. 5 4 Specifi- cally,in the next section I willimplement an analysis according to which - t-/-s-arethe default instantiationsof a functionalhead found in ‘ ‘deverbal’’ syntacticobjects. In doing this, I will usethe abbreviation Default-Asp torefer totheconditions under which the exponents - t-/-s- are realizedon Asp.

6.4Deverbal Formations Thereare two questions that must be addressedin analyzing the ‘ ‘pastpassive participle,’ ’ based ontheidea that the participle is a typeof ‘‘deverbal’’ formation.First, the status of anotionlike ‘‘deverbal’’ mustbe capturedin a theorywithout categories like ‘ ‘V’’inthe first place. Second, theactual realization of the perfect with the affixes - s- or -t-mustbe addressed. The analysis thatI presentlinks the deverbal status of the relevant forms directlyto the behavior of the exponents -t-/-s-.Whatit means for somethingto be deverbal in this sense is that it appears beneaththe functional head Asp (and perhaps v aswell), but not combined with T. Themotivation for positingthe head Asp in suchcases lies in thesyntax and semantics as well, in thataspectual propertiesare implicated in nominalizations and other deverbal formations. The affixes - s- and -t-arethen the default realization of this aspectual head, and this accounts for theirdistribution. Tobegin with, realization as the morphological object V amountsto combining (through headmovement or Merger) with T. Thiscombination could fail to occur for morethan one reason.

52 Someattempts havebeen made toaccount for this extension of the- s-form;thus, for instance, Sommer (1914: 607)suggests that the presence ofverbs with perfects in- s- (e.g., scrib-o¯ ‘write’, perf. scrip-s-¯õ)facilitated thespread of the -s-participle. 53 As Aronoffnotes, there are exceptionsto this, but not for - t-/-s-;rather,there are afew verbsin which the phonologyof the Root `Theme preceding- t-/-s-is differentin the two cases. Thus,for instance, the (deponent) verb morior ‘die’has as pastparticiple mor-tuus, as futureactive mor-i-t-u¯r-us. Althoughthe - t-allomorphappears inboth cases, thetwo participles show other differences. 54 Aronoffderives further conclusions from this point that I donotaccept; see section8 forsome discussionof his framework. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 217

Innominalizations T willnot be present, 5 5 rather,the functional head D willultimately dominate theposition of the Root (although Asp will be present). In the perfect with the feature [pass], therewill be nocombinationof the Ï -v-Asp complexwith T. Effectively,then, a deverbalnoun oradjective is a Rootappearing in the structure in (43), when this structure does not combine withT (althoughother functional heads might be present).Specifically, I proposethat deverbal formationsshare as a subcomponentthe piece of tree shown in (43). (43) Common structure of deverbal formations AspP

Asp vP

v Ï P

Ï (DP) Recallthat the head v hereis the functional ‘ ‘lightverb’ ’ andthat the head Asp contains featuresrelating to perfectivityand imperfectivity. Taken together, the two functional heads here containbasic information about eventivity /stativity,along with further aspectual information about thestatus of the event or state. Depending upon which functional heads appear higher up in the structure,the resulting object will be an ‘ ‘A’’oran‘ ‘N,’’ ifitdoes not combine with T. 5 6 Inthis structure - t-/-s-aredefault instantiations of thehead Asp. 5 7 Thisfollows the first part ofthe analysis presented in Aronoff 1994; - t-/-s-haveno consistent content in terms of voice, aspect,and so on. Rather, the common property of forms withthese exponents consists in their simplybeing deverbal, and not something more specific than this. When there is a morehighly specifiedvocabulary item with a properset of features, such as thepresent participle in - nt- (cf. laud-a¯-ns ‘praising’), 5 8 thiswill prevent insertion of thedefaults (see (44)). Thereare two further components to analyzingthe behaviorof - t-/-s-intermsof thisstructure. Thefirst has to dowithwhat determines the choice between these two for aparticularverb. This issimply a listedproperty of these vocabulary items, unconnected with deeper generalizations. 5 9 TheAsp node in thetree will always be specifiedfor featureslike [perf] and[pres]; butfor the

55 Or willbe dominated by D, and affected accordingly.Certain approaches do in fact positT withinthe DP. 56 Forinstance, in the nominalizations (e.g., the supine) a Dwillbe present. 57 Thedefault status of ‘‘perfect’’ participialmorphology also surfaces inIppolito’ s (1999)analysis of Italian and Arregi’s (1999)analysis of Basque,suggesting further crosslinguistic questions that I cannotpursue here. 58 The -nt-suffixis realized inthe(nominative singular) citation form as - ns. 59 As notedearlier, allomorphybetween - t- and -s-is notphonologicallypredictable, although it was originally(see, e.g.,Sommer 1914,Ernout 1952 /1989);hence thelist here. Having an - s-as opposedto - t-inthis object does not correlate withother types of allomorphy (e.g., in thesynthetic perfect). Thus,there are forms with- s-inthe synthetic perfect, like scrib-o¯ ‘write’, perf. scrip-s-¯õ, with -t-participles: scrip-t-us ‘written’. At thesame time there are verbswith - s-participles, but not -s- perfects: fall-o¯ ‘deceive’, perf. fe-fell-¯õ, participle fal-s-us. 218 DAVIDEMBICK realizationof Default-Asp,what is requiredis underspecification of thevocabulary items, such that the -t- or -s-willappear when no more specific instantiation of Aspcan be inserted.Thus, atleastthe following vocabulary items will be relevant, where the conditions for theinsertion of thepresent participle - nt-affixare given for purposesof exposition: (44) Realizationof Asp (not raised to T) -nt- [pres] $ -s- [ ] / (List) $ -t- [ ] $ Thesecond component has to do with adjustments to the stem of the verb that take place inthe deverbal - t-/-s-forms. Ina numberof cases the phonology of the stem of the verb is differentin present, perfect, and past participle environments. Thus, for instance,in the ‘ ‘nasal- infixing’’ class,there are sets like tang-o¯ ‘touch’with perfect te-tig-¯õ, participle ta¯c-t-us. On the viewI amassuminghere, stem allomorphy of thistype is distinct from affixation.It results from whatHalle and Marantz (1993) classify as readjustmentrules, morphophonologicalprocesses thatapply to alter the shape of stems. In particular, the point is that the certain phonological readjustmentsto thestem could be tiedto the presence of - t-/-s-.Whena morehighly specified affixwins out over - t-/-s-,thereare no apparent effects on thephonologyof the stem. In traditional terms,this is expressed by sayingthat certain participial forms (e.g.,the present, the gerundive, gerund)are based on the‘ ‘presentstem.’ ’ Ultimatelythis may or may not be the result of there beingphonological effects associated with consistent sets of features in the environment of the stem.6 0 Summarizingthe approach, the conclusion is that - t-/-s-arethe default realization of anAsp head,which is a componentcommon to the various deverbal structures. 6 1 Theappearance of theseexponents in both the ‘ ‘pastpassive’ ’ and‘ ‘futureactive’ ’ participlespresents no contradic- tion;each of thestructures associated with these two participles contains an Asp head, which is realizedwith the default affix. In the case of thefuture active, I assumethat additional structure ispresent,pertaining to the modal nature of the interpretation ‘ aboutto ...’. 6 2 Thetwo structures, shownwith the relevant exponents in the positions that they instantiate, are given in (45) and (46).

60 Forinstance, the present and perfect systems couldbe dividedalong Varro’ s distinction infectum versus perfectum (see Varro 1938;essentially imperfective versusperfective, but see Binnick1991 for discussion and a critiqueof this classification).This would accord withthe fact thatstem suppletionis conditioned by this distinction, in fer-o¯ ‘carry’ in imperfective tense /aspects, tul-¯õ inthe (synthetic) perfect. Itshould also be noted that adjustments to the stem are associated withthe synthetic perfect; see footnote7. 61 Thesyntactic analysis discussed to this point involves two distinct heads, v andAsp, which are cruciallyinvolved indefining deverbal status. In a numberof cases aspectual specificationand the licensing of agentsare intimatelyconnected withone another. For instance, the generalizations associated withthe distinction between ‘‘adjectival’’ and‘ ‘verbal’’ passives made, forexample, by Wasow (1977)are largelyrestatable solelyin terms ofaspect: stativeversus eventive (see Pesetsky1995, Marantz 1995).How these connectionsbetween aspect andagentivity /etc. are tobe captured is an interestingquestion, but one I cannotaddress here. 62 Itake itas uncontroversialthat modal information appears indeverbal structures of thistype; witness, for example, theso-called gerundive of Latin: from laud-o¯ ‘praise’, there is thegerundive laud-a-nd-us ‘(one)to bepraised’ , withthe necessitative interpretationof English ‘ ‘modal be’’ (Johnis tobe praised ). FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 219

(45) “Past passive” (46) “Future active” (example: am-a¯ -t-us ) (example: am-a¯ -t-u¯ r-us) AspP ModP

Asp vP Mod AspP

-t-/-s- v Ï P -u¯ r- Asp vP

Ï DP -t-/-s- v Ï P

Ï DP Theexamples here indicate the surface forms thatwill be realizedfollowing head movement of the Ï -v toAsp or Modin each of thesecases. Furtherquestions, concerning what precisely makes the second of thetwo participles above activeas opposed to passive syntactically, are part of a muchlarger set of issues concerning correlationsbetween voice, aspect, and syntactic structure that I cannothope to addresshere (cf. footnote61).

6.5Other Aspects of Default-Asp Inaddition to therebeing participles in - t-/-s-formedfrom Rootswith verbal forms, thereare those thatapparently take the ‘ ‘participial’’ ending- t-(Allenand Greenough 1931:149, Gildersleeve and Lodge1895:130; also see Joffre 1986),but for whichno verbal forms arefound; that is, these Rootsdo not appear as verbs, combined with T. 6 3 (47) fu¯nus‘ death’: fu¯nes-t-us‘ deadly’ honor‘ honor’: hones-t-us‘ honorable’ barba‘ beard’: barb-a¯-t-us‘ bearded’ turris‘ tower’: turr- ¯õ -t-us‘ turreted’ cornu¯ ‘horn’: corn-u¯-t-us‘ horned’ Theinterpretation is somethinglike ‘ providedwith’ . Unlikewhat we foundwith the participles in -s-/-t-, however,there are no verbal forms for theseRoots, which are otherwise realized in nominalenvironments. This connects with a furtherfact about ‘ ‘perfectpassive participles.’ ’ Althoughthe participles are in many cases interpreted as having perfect /perfectiveaspectual

63 Because there are noverbalforms, we cannotestablish the abstract identityunderlying this affix as we couldwith Default-Asp- t-/-s-inthe deverbal formations discussed above. 220 DAVIDEMBICK properties,this is not always the case; thus, Brugmann (1895:secs. 4, 7) observes that - t-/-s- adjectivesappear with both present and perfect interpretations. 6 4 (48)a. maereo ¯ ‘besad, grieve’ , maes-t-us‘ fullof sadness,sorrowful’ caveo¯ ‘beon one’ s guard’, cau-t-us‘ cautious’ taceo¯‘besilent’, taci-t-us‘ silent’ b.vir lauda ¯tus‘ manbeing praised’ filius u¯nice¯ama¯tus‘ especiallyloved son’ virab omnibus contemptus ‘ mandespised by all’ regio¯ habita¯ta‘ inhabitedregion’ The‘ ‘dualfunction’ ’ ofthe same form isexplained by Brugmann (1895:94) as resulting from thefact that both are essentially predicates and that the ‘ ‘adjectival’’ useof theform denoted acharacteristicproperty, which was, likethe participial use, in somesense complete. Joffre (1986) seemsto begetting at something similar in identifying- t-adjectivesand participles as describing astatethat is ‘ ‘nonde ´passe´’’(‘not passed’ ; i.e.,presumably one that simply holds). Given the status of -t-/-s-asdefault realizations of Asp, these facts are less surprising. Stativization as construedin a broadsense can apply to eventiveverbs, stative verbs, and ‘ ‘nouns’’ alike.In the caseof the derivation from theRoots in (47),the idea is that to beprovidedwith a qualitydenoted bythe Root is in some sense an aspectual notion. Thus, the fact that ‘ ‘participial’’ morphology shouldappear on Roots that do not appear as finite verbs becomes somewhat less obscure. Al- thoughthe question of thenonoccurrence of suchforms asverbsis stillopen, the discussion of thissection points to adirectconnection between the properties of Aspand the properties of the functionalstructure associated with ‘ ‘adjectives.’’6 5 Theabsence of verbalforms mightthen stem from thefact that stativizing Asp, spelled out as - t-,appliessyntacticosemantically in a wide rangeof environments. Thatis, there are features of Aspthat function quite generally on a syntacticosemanticlevel asstativizers;but these are always realized with the default realizations of Asp - t-/-s-, even when thefeatures apply to different complements. This would account for thevariable interpretation ofthe‘ ‘participles,’’ aswellas thebehavior with Roots that do notform verbs.The absence of verbalforms inthe relevant cases would then be acceptable; the appearance of a Rootin the stativizedstructure does not necessarily imply that the same Root will function independently as a‘‘verb.’’ Butthe conclusion must remain speculative, as there is in the case of the Roots in (47)no argumentfor themorphological identity of the- t-withthat found in deverbal formations. 6 6

64 This,for Brugmann, was thedifference between adeverbaladjective, which is aspectuallyindifferent, and a participle,which is aspectuallyspecified. The aspectual ( 4 Aktionsart)propertiesof the underlying verbs also play a role,as Hofmann,Leumann, and Szantyr (1963:391) note. 65 Forthe question of whatmight underlie the apparent gaps of this type, see Ippolito’s (1999)analysis of Italian participialformations. 66 Theoverlap between denominaland deverbal adjectives mightbe found in English as well,with denominal adjectives meaning‘ providedwith’ formed with ‘ ‘participle’’ endings. (i)a. beard:bearded horn:horned wing: winge`d FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 221

7Comparisonwith LexicalistTreatments Inowaddress possible Lexicalist and syntactic approaches to the phenomena analyzed above. The term Lexicalist hererefers totheoriesof thetype developed in Lieber1980, Kiparsky 1982, Di Sciulloand Williams 1987, and related work, whose defining property for presentpurposes isthat affixation in apresyntacticlexicon determines syntactic behavior. That is, on thistype of theoryaffixes exist in a lexiconlike everything else; the properties for whichan individual affixis specified determine the behavior of complex words through mechanisms governing the inheritanceor ‘‘percolation’’ offeatures.Thus, to take a verysimple example, the English plural morphemewould consist of thefollowing, along with a specificationof the phonological shape (viz., /-z/):

(49) [[N ][-z [pl]]] The[pl] feature becomes a featureof theentire derived word, now a pluralnoun. The same type ofconsiderationapplies to more complex cases, for example,passives. The formation of apassive verbis directly related to affixation with ‘ ‘passivemorphemes,’ ’ andthisproduces passive syntax. Tobegin with, a Lexicalistapproach to deponentverbs faces a verygeneral problem, even withoutsomething like the Latin analytic /syntheticsplit. This is discussed in detail in Embick 1997,1998; I willsummarize the major points of those treatments here. The major objection concernshow deponents would have to be specifiedon a Lexicalisttreatment in the first place. Ona Lexicalistapproach affixation determines the syntactic behavior of theaffixed word. Affixa- tionof passivemorphology will result in verbsthat may appear only in passive syntactic environments;that is, in this type of theory affixation of passive morphology will, in effect, force a passivesyntactic configuration. Deponent verbs have passive morphology. But deponent verbs ofthetype discussed above do nothave passive syntax. Thus, the morphology found in deponents andthat found with passive verbs cannot be the same in terms of its morphosyntactic features. Oneset has to produce passive syntax; the other set has to be simply associated with a lexical quirk.Yet this misses significant generalizations. Passives and deponents are morphologically identicalthrough all tenses, persons, and so on, and moreover behave the same way with respect tothe formation of theperfect. In a Lexicalisttheory, however, the most that can be saidis that thereare two distinct sets of affixesfound with passives and with deponents, which, despite being

b.tooth: saber-toothed hair:long-haired tail:bushy-tailed The(b) examples are onlypossible with further modification (i.e., * haired);see Marantz 1989for discussion of similar cases. As far as thisconnection goes in English, there isa furthercomplication. In some cases thedenominal adjectives with -ed requirea syllabicpronunciation of the suffix: thus, winge`d in(ia). Dubinsky and Simango (1996) note that the reasonfor this lies inthe fact thathistorically two distinct - ed suffixesare involvedhere. They note further that for certain verbs,there are twopronunciations of the- ed suffixedforms: for example, /blest/ and /blesse`d/ for bless. Theformer is forthem averbalpassive, the latter adjectival.The main thrustof their argument concerning these forms is directed againstthe position taken in, for example, Levin and Rappaport 1986 (and earlier work)that adjectival passives are zero- derivedin the lexicon from verbal passives. However, many instances ofthistype appear tobe adjectives withoutthe aspectual force ofthe‘ ‘adjectival’’ passive,rendering the status of Levin and Rappaport’ s argumentunclear. 222 DAVIDEMBICK phonologicallyidentical in all cases, have distinct morphosyntactic features. The identity of the twotypes of verbs,captured here with the single feature [pass], is completely missed on thistype oftreatment. Havingnoted the above, I shouldpoint out that the behavior of deponentverbs could always bestated formally in a Lexicalisttheory. It could be argued, for instance,that although certain typesof ‘ ‘passivemorphology’ ’ propagatefeatures associated with passive syntax (or project ‘‘passivesyntax’ ’), thesame is not true of passivemorphology with a lexicallyspecified set of verbs,the deponents. In formal terms this is, of course,a possibility.The point is that this type ofanalysis results in a stipulationin a componentof the grammar thatdetermines syntactic behavior.On this type of theory, the behavior of complex words is determined in the lexicon; thefact that phonologically identical sets of affixes have radically different syntactic effects can thenonly be captured as a stipulation,and one that strips away the idea that lexical processes determinesyntax. On the other hand, treating voice features like [pass] asidentical in bothcases, ason the present account, captures the relevant syntactic effects (analytic vs. synthetic forms), withoutlosing the identity of the feature in both passive syntax and deponent verbs.

8Syntaxand Paradigms 8.1Paradigmatic Perspectives on Syntax and Morphology Thesplit behavior of the Latin perfect might be seenas supportinga ‘‘paradigmatic’’ approach tomorphology. Paradigmatic approaches are proposed by Matthews (1972) and developed and elaboratedin Anderson’ s (1982,1992) ‘ ‘extendedword-and-paradigm’ ’ and‘ ‘a-morphous’’ approaches.The most relevant property of theoriesof thistype is thatthey allow for agreatamount oflaxityin the interface between syntax and morphology; syntactic distinctions and positions are potentiallyobliterated on the way to morphological realization. 6 7 Thereason for thisis that ‘‘morphemes’’ haveno statusin theories of thistype, beyond being the by-products of morpho- phonologicalrules. Associating specific signals with specific syntactic objects in a hierarchical structureis thusnot possible. On the paradigmatic approach, sets of morphosyntacticfeatures are associatedwith rule blocks. The rules are conditioned by the features, and any phonological materialinserted onto a form ismerelya by-productof theoperation of the abstract rule. Because phonologicalpieces are simply by-products of rules, rather than the exponents of fixed positions ina structurebuilt by the syntax, the possibility exists that morphological realizations will bear littlerelation to each other or tothesyntactic structures in which these forms appear.The clearest exampleof thistype of theory for presentpurposes is inthe ‘ ‘morphology-by-itself’’ approach,

67 There is alsoa typeof paradigmatic approach to morphology, exemplified in the work of Carstairs-McCarthy, thatdiffers from this type in many respects. Carstairs-McCarthy(1987, 1992) recognizes that an important goal of any morphologicaltheory is tostate andcapture the constraints on deviation from one-to-one patterns in the relationship between syntax /semantics andmorphology. In his approach the objects over which such constraints are stated are entire paradigms,that is, entire sets ofinflected forms oflexemes. Thus,irrespective ofhowthis ‘ ‘paradigm’’ relates tosyntax, constraintsare notstated at thelevel of features andtheir interactions, but on sets ofinflected forms arrangedin acertain way andconsidered as aprivilegedtheoretical object. Bo ¨rjars, Vincent,and Chapman’ s (1996)treatment discussedbelow appeals toanextrasyntactic notion of paradigm as well. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 223 elaboratedby Aronoff (1994).Aronoff’ s theoryposits a ‘‘fullymediated’ ’ syntax /morphology interaction.An intervening level exists between syntax and morphological realization, such that the differencesbetween disparate syntactic objects can be completelyremoved prior to morphological realization(see Aronoff1994:25). 6 8 Theexistence of corresponding analytic and synthetic forms inLatin,along with the existence ofdeponentverbs, seems to pointto discrepanciesbetween syntax and morphology: in one case anunexpected difference in composition, in the other a problemwith ‘ ‘passive’’ morphology withoutpassive syntax. A theorywith an intervening level mediating syntax and morphology, whichsyntactic structures can be related to arbitrarily, is able to accommodate this apparent mismatchquite easily. 6 9 Theproblem with the paradigmatic view in thepresent case is thatit can in principle make nopredictions about what is realized in the analytic perfect. On this type of treatment, a - t-/-s- participleis realized because a particularset of features happens to be associated with this object—thatis, with a specificrealizational rule, or with a particular‘ ‘stem,’’ asin Aronoff’ s treatment.But the analysis presented above shows that what is found in the analytic perfect followspredictably from asyntacticstructure that is thesame as theone underlying the synthetic perfect.The distribution of - t-/-s-,andits presence in certainforms butnot others, follows from thedefault nature of these vocabulary items and from astructuralanalysis that is motivated by syntaxand semantics. The analysis furthermore carries over to therange of deverbal formations inwhich the - t-/-s-affixesappear. There is a separationbetween syntax and morphology on the approachadvocated here. But the default case is for themorphology to instantiate terminals in syntacticstructures, thus ensuring direct interaction. Paradigmatic theories like that proposed by Aronoff,on theother hand, emphasize this separation to the extent that direct interactions are at besta convenientaccident. 7 0 Thatis, the worst possible case, with erasure of syntacticdistinctions inthe morphology, is the norm, or at least not differentiated from thenorm. Thereis a secondvariety of paradigmatic approach to syntax /morphologyinteractions, in whicha ‘‘paradigmspace’ ’ generatedoutside of the syntax contains entries in cells that are correlatedwith morphosyntactic feature combinations. A specificparadigmatic treatment, part of whichis directed specifically at the two forms oftheLatin perfect, illustrates a numberof points particularlywell. Bo ¨rjars,Vincent, and Chapman (1996), assuming that the two forms ofthe perfectmust fill slots in the same paradigm, base their analysis on the idea that positions in a paradigmmay be filled by syntactically complex objects. In essence this analysis makes the followingclaims: first, that there is an extrasyntactic space, the ‘ ‘paradigm,’’ containingforms

68 Theloss of informationbetween syntaxand morphological realization in this type of theoryis distinctfrom the neutralizationof features inthemorphology, that is, impoverishment in the sense ofBonet1991 and subsequent work (see, e.g.,Noyer 1992, 1998, Halle 1997).Impoverishment does not eradicate syntacticstructures, only particular features. Thepossibility for complete erasure ofsyntactic distinctions thus does not exist. 69 Inprinciple it is capable ofaccommodating almost everything,subject perhaps only to considerations of learnability. ButI willconcentrate here onillustratingthe difficulties presented by theperfect forthis type of theory. 70 Inother paradigmatic theories the separation would not be as extreme; thefocus here ison Aronoff’ s treatment forexpository purposes. 224 DAVIDEMBICK correspondingwith feature values; and, second, that the objects in the paradigm may be syntacti- callycomplex, that is, have to be regardedas constructed by thesyntax. With these two properties, thisis a theoryin which any relationship between syntax and morphology can simply be listed intables; there are no interesting predictions to be made, because absolutely anything could fill acellin aparadigm:any ‘ ‘word,’’ anyphrase, any sentence. Regular realization and suppletion areon exactlythe same plane. 7 1 As aresult,the theory can in principlemake no predictions about possiblesyntax /morphologyconnections. Although it is true that the occurring forms oftheLatin perfectcan be described in this notation, the account offers nothingbeyond a recitationof the basicfacts.

8.2Syntax and Separation Theanalysis I havepresented above of theLatin perfect is, in its essence, syntactic. But treating theLatin perfect syntactically does not amount to ignoring the fact that it isaspecialcase. There is,of course, something odd in this system. Owing to properties of Latin morphology, isolated inthefeature [pass], the two types of perfect differ from eachother on thesurface. This is simply avagaryof Latin, the origin of which is ultimately diachronic. Yet despite this quirk, the two forms arenot completely unrelated to oneanother. That is, the components of the analytic perfect resultpredictably in a perfectpassive (or ina deponent,a perfect).On the paradigmatic view, thatthis should be the case is in no way ensured. Owing to properties of its specification, the ‘‘pastperfect participle’ ’ doesnot contain vocabulary items that specifically contain the features [perf] and[pass]. This is due to properties of thespecification of - t- and -s-,thatis, their default distribution.But their appearance is predictable given the structure that underlies the analytic perfect.Even with the differences in surfacerealization, the forms foundin the perfect are consistent;and the consistency follows from asyntacticanalysis. Havingnoted the above, I mustacknowledge that the Latin system represents a double departurefrom adirectmodel of syntax /morphologyinteractions, where ‘ ‘direct’’ refers tothe simplecase in whichexponents are inserted into terminal nodes arranged by thesyntax, and that istheend of the story.First, the voice system involves features like [pass], whose status is unclear. Second,the existence of deponentverbs represents a furtherdeparture, in which the same feature [pass],which is inonecase systematically related to the syntax, is also associated inherently with certainRoots. But these departures are deviations from thenormal state of affairs. Inthe treatment Ihaveprovided, the points of departure from thedirect model are isolated and constrained, not generalizedto overarching principles. The connections between structure and form inthe Latin perfectmay be complicated, but they are not random. At theheart of the matter, the syntactic andsemantic identity of the structure resulting in thetwo perfects provides the basis common to thetwo forms.

71 See Bo¨rjars, Vincent,and Chapman 1996:159, where itisnoted that in terms oftheiranalysis ‘ ‘...asuppletive formand a regularlyderived form have exactly parallel feature representations.. ..Therefore,as far as thegrammar is concerned,they behave indistinguishably from each other.’’ FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 225

Ona moregeneral level, the comparison of the paradigmatic and syntactic approaches to syntaxand morphology suggests how theories of thesyntax /morphologyinterface are to becon- strained.All of the theories discussed in this section accept the SeparationHypothesis, theidea that thereis a distinctionbetween morphosyntactic features and the morphophonological exponents associatedwith them. The move to separation is required in order to account for basicfacts aboutsyntax /morphologyinteractions —inparticular, for systematicneutralizations of features, requiringunderspecification, and default vocabulary items. I willnot trace the history or thefull motivationbehind this idea here; Beard 1995 contains relevant background discussion. The point isthat a burdencomes with this move. Separating syntax and morphology in this way raises questionsabout how and why the interaction between these two components is primarily direct—that is, why there are, for themost part, systematic correlations between syntax and morphol- ogyin the first place. The assumption that has driven this investigation is that the interface is constrainedbecause, in the default case, morphology interprets syntactic structures directly.

9Conclusion Interms of what it presents for thestudy of verbsand participles, and for thestudy of category moregenerally, the Latin perfect is aspecialcase. This is due to itsmixed nature. As aresultof this,it canbe shownthat analytic and synthetic forms derivefrom thesame syntactic structure. Whatcan be shownin thisparticular case contrasts with other cases in which one findsan auxiliary anda participle.In any auxiliary `participleformation, one could argue that the participlecontains information(aspectual, etc.) not associated with regular adjectives. This much is clear. But it doesnot specify the manner in which the ‘ ‘adjectivalization’’ ofverbs occurs. Nor, for that matter,is the relationshipbetween the auxiliary `participleforms andother synthetic verbal forms clarified.The importance of this is that, in such cases, it could always be claimed that analytic andsynthetic forms differradically in their syntactic structure, or inthemorphosyntactic features involvedin the two forms. Theimportance of theLatin perfect is that it presentsa casein which thefeature differentiating analytic from syntheticforms canbe isolatedand examined, and, as a resultof this,it can be shown that the two types arise from thesame syntactic structure. Aclearimplication of the discussion presented here is in the domain of category. On the treatmentof the perfect forms thatis provided above, notions like morphological ‘ ‘adjective’’ and‘ ‘verb’’ areepiphenomenal: they result from generalproperties of thelanguage, statable in termsof syntactic structures. This does not mean that there are no syntacticcategories; rather, it meansthat there is no simple and straight correspondence between lexical category labels like N,V,andA onboth syntactic and morphological levels. What other features Roots may be specifiedfor isstill an open question. In any case the analysis highlights the nature of part-of- speechlabels by essentially dispensing with them. However, the treatment developed here is restrictedto deverbal formations, where the effects of ‘ ‘category-changing’’ morphologyare reducedto syntactic structures and postsyntactic morphological processes. The broader question, concerningwhether all cases of derivationalmorphology can be analyzedwithout appeal to simple connectionsbetween syntactic categories and morphological category labels, is amatterfor further research. 226 DAVIDEMBICK

Theanalysis also points to aconnectionbetween Roots and a featurethat plays a rolein the syntax.The implication of thispart of the discussion is thata simplemodel of syntax /morphology interactions,in which Roots cannot be associated with syntactic features inherently, cannot be maintained.What remains to be seen is if this case fits with others crosslinguistically. For the timebeing, however, it provides a clearinstance in which features associated arbitrarily with certainRoots have an effect that is not merely morphological. This points to a furtherset of questionsregarding the hypothesis that Roots are subject to LateInsertion into syntactic structures, andregarding the relationships between features in syntax and morphology, questions that cut acrossseveral components of the grammar. One of thesolutions advanced above is one in which thefeatures of certainRoots are visible in thesyntactic derivation, and in whichthis is sobecause certainRoots must be present in the syntax to beginwith. Whether or not this solution is the best onedepends upon the resolution of a hostof issues that are only now becoming clear. The suggestionin light of this was thatin order for theproperties of Roots that are visible in the syntaxto be identifiedexactly, the working hypothesis should be that no Root features are syntacticallyactive, unless an argument to the contrary can be produced.

References Allen,Joseph H. 1931. Allenand Greenough’ s newLatin grammar for schools and colleges. New York: Ginn. Anderson,Stephen. 1982. Where’ s morphology? LinguisticInquiry 13:571–612. Anderson,Stephen. 1986. Disjunctive ordering in inflectional morphology. NaturalLanguage & Linguistic Theory 4:1–32. Anderson,Stephen. 1992. A-morphousmorphology. Cambridge:Cambridge University Press. Aronoff,Mark. 1994. Morphologyby itself: Stems and inflectional classes. Cambridge,Mass.: MIT Press. Arregi,Karlos. 1999. Tense in Basque. Ms., MIT, Cambridge, Mass. Baldi,Phillip. 1976. Remarks on the Latin r-form verbs. Zeitschriftfu ¨rvergleichendeSprachforschung 90: 222–257. Beard,Robert. 1966. The affixation of adjectivesin contemporaryliterary Serbo-Croatian. Doctoral dissertation,University of Michigan,Ann Arbor. Beard,Robert. 1995. Lexeme-morphemebased morphology: A generaltheory of inflectionand word formation. Albany:State University of New YorkPress. Belletti,Adriana, and Luigi Rizzi. 1988. Psych-verbs and u -theory. NaturalLanguage & LinguisticTheory 6:291–352. Benveniste,E ´mile. 1948. Nomsd’ agent et noms d’ action en Indo-Europe ´en. Paris:Adrien-Maisonneuve. Binnick,Robert I. 1991. Time andthe verb. Oxford:Oxford University Press. Bobaljik,Jonathan. 1994. What does adjacency do? In MITworkingpapers in linguistics 21: The morphology- syntaxconnection, 1–32.MITWPL, Departmentof Linguistics and Philosophy, MIT, Cambridge, Mass. Bobaljik,Jonathan. 1995. Morphosyntax: On the syntax of verbal inflection. Doctoral dissertation, MIT, Cambridge,Mass. Bonet, Eula`lia.1991. Morphology after syntax: Pronominal clitics in Romance. Doctoral dissertation, MIT, Cambridge,Mass. Bo¨rjars,Kersti, Nigel Vincent, and Carol Chapman. 1996. Paradigms, periphrases, and pronominal inflection: Afeature-basedaccount. Yearbookof Morphology1996, 155–180. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 227

Brugmann,Karl. 1895. Die mit dem Suffix -to- gebildetenPartizipia im Verbalsystem des Lateinischen und desUmbrisch-Oskischen. IndogermanischeForschungen 5:89–153. Carstairs,Andrew. 1987. Allomorphyin inflexion. London:Croom Helm. Carstairs-McCarthy,Andrew. 1992. Morphologicaltheory. London:Routledge. Chantraine,Pierre. 1927. Histoiredu parfait grec. Paris:H. Champion. Chomsky,Noam. 1970. Remarks on nominalization. In Readingsin English transformational grammar, ed. RoderickJacobs and Peter Rosenbaum, 184– 221. Waltham, Mass.: Ginn. Chomsky,Noam. 1995. TheMinimalist Program. Cambridge,Mass.: MIT Press. Chomsky,Noam. 1998. Minimalist inquiries: The framework. (MIT Occasional Papers in Linguistics 15.) MITWPL, Departmentof Linguistics and Philosophy, MIT, Cambridge, Mass. [To appear in Step by step, ed.Roger Martin, David Michaels, and Juan Uriagereka. Cambridge, Mass.: MIT Press.] Di Sciullo,Anna Maria, and Edwin Williams. 1987. Onthe definition of word. Cambridge,Mass.: MIT Press. Draeger,Anton. 1878. HistorischeSyntax der lateinischen Sprache. Leipzig:B. G.Teubner. Dubinsky,Stanley, and SilvesterRon Simango. 1996. Passive and stative in Chichewa: Evidence for modular distinctionsin grammar. Language 72:749–781. Elbourne,Paul. 1999. Some correlations between semantic plurality and quantifier scope. Paper presented atNELS 29[toappear in theproceedings]. Embick,David. 1997. Voice and the interfaces of syntax. Doctoral dissertation, University of Pennsylvania, Philadelphia. Embick,David. 1998. Voice systems and the syntax /morphologyinterface. In MIT workingpapers in linguis- tics32: Papers from the UPenn /MIT Roundtableon Argument Structure and Aspect, 41–72. MITWPL, Departmentof Linguistics and Philosophy, MIT, Cambridge, Mass. Embick,David, and Morris Halle. 1999. The Latin conjugation. Ms., MIT, Cambridge, Mass. Embick,David, and Rolf Noyer. 1999. Locality in post-syntactic operations. In MITworkingpapers in linguistics34: Papers in morphology and syntax, cycle two, 265–317. MITWPL, Departmentof Linguisticsand Philosophy, MIT, Cambridge, Mass. Ernout,Alfred. 1952 /1989. Morphologiehistorique du latin. 4thed. Paris: Klincksieck. Ernout,Alfred, and Franc ¸oisThomas. 1951. Syntaxelatine. Paris:Klincksieck. Flobert,Pierre. 1975. Lesverbes de ´ponentslatins des origines a` Charlemagne. Paris:Belles Lettres. Gellius,Aulus. 1927. TheAttic nights of AulusGellius, with an Englishtranslation by JohnC. Rolfe. London: W.Heinemann. Gildersleeve,Basil, and Gonzalez Lodge. 1895. Latingrammar. London:Macmillan. Giorgi,Alessandra, and Fabio Pianesi. 1991. Toward a syntaxof temporal representations. Probus 3:187–213. Giorgi,Alessandra, and Fabio Pianesi. 1997. Tenseand aspect: From semantics to morphosyntax. Oxford: OxfordUniversity Press. Glare,P. G.W.1982. OxfordLatin dictionary. Oxford:Oxford University Press. Gonda,Jan. 1960a. Reflections on the Indo-European medium I. Lingua 9:30–67. Gonda,Jan. 1960b. Reflections on the Indo-European medium II. Lingua 9:175–193. Hale,Kenneth, and Samuel Jay Keyser. 1993. On argument structure and the lexical expression of syntactic relations.In Theview from Building 20: Essays in linguistics in honor of Sylvain Bromberger, ed. KennethHale and Samuel Jay Keyser, 53 –109.Cambridge, Mass.: MIT Press. Halle,Morris. 1990. An approach to morphology. In NELS 20, 150–184.GLSA, Universityof Massachusetts, Amherst. Halle,Morris. 1997. Distributed Morphology: Impoverishment and fission. In MITworkingpapers in linguis- tics30: Papers at the interface, 425–449.MITWPL, Departmentof Linguistics and Philosophy, MIT,Cambridge, Mass. 228 DAVIDEMBICK

Halle,Morris, and Alec Marantz. 1993. Distributed Morphology and the pieces of inflection. In The view fromBuilding 20: Essaysin linguisticsin honorof SylvainBromberger, ed.Kenneth Hale and Samuel JayKeyser, 111 –176.Cambridge, Mass.: MIT Press. Halle,Morris, and Alec Marantz. 1994. Some key features of Distributed Morphology. In MIT working papersin linguistics 21: Papers on phonology and morphology, 275–288.MITWPL, Department ofLinguisticsand Philosophy, MIT, Cambridge, Mass. Harley,Heidi. 1995. Subjects, events, and licensing. Doctoral dissertation, MIT, Cambridge, Mass. Harley,Heidi, and Rolf Noyer. 1998. Licensing in thenon-lexicalist lexicon: Nominalizations, vocabulary itemsand the encyclopaedia. In MITworkingpapers in linguistics 32: Papers from the UPenn /MIT Roundtableon ArgumentStructure and Aspect, 119–138. MITWPL, Departmentof Linguisticsand Philosophy,MIT, Cambridge, Mass. Hofmann,Johann Baptist, Manu Leumann, and AntonSzantyr. 1963. LateinischeGrammatik, auf der Grund- lagedes Werkes von Friedrich Stolz und Joseph Hermann Schmalz: 2. BandLateinische Syntax und Stilistik. Mu¨nchen:Beck’ sche Verlagsbuchhandlung . Hovdhaugen,Even. 1987. Generaverborum quot sunt? Observationson the Roman grammatical tradition. In Thehistory of linguistics in theclassical period, ed.Daniel J. Taylor,133 –147.Amsterdam: John Benjamins. Iatridou,Sabine. 1990. About Agr(P). LinguisticInquiry 21:551–577. Ippolito,Michela. 1999. On the past participle morphology in Italian.In MIT workingpapers in linguistics 34:Papers in morphology and syntax, cycle one, 111–138.MITWPL, Departmentof Linguistics andPhilosophy, MIT, Cambridge, Mass. Joffre,Marie-Dominique. 1986. La signification temporelle et aspectuelle d l’adjectif en - *to: Re´flexions sur l’inte´grationd’ uneforme adjective au paradigme verbal. Revue des E´tudesLatines 64:211–222. Kemmer,Suzanne. 1993. Themiddle voice. Amsterdam:John Benjamins. Kemmer,Suzanne. 1994. Middle voice, transitivity, and the elaboration of events. In Voice:Form and function, ed.Barbara Fox and Paul J. Hopper,179 –230.Amsterdam: John Benjamins. Kiparsky,Paul. 1982. Lexical morphology and phonology. In Linguisticsin themorning calm, ed.Linguistic Societyof Korea,3– 91. Seoul: Hanshin. Klaiman,Miriam H. 1991. Grammaticalvoice. Cambridge:Cambridge University Press. Kratzer,Angelika. 1993. The event argument and the semantics of voice. Ms., University of Massachusetts, Amherst. Leumann,Manu, Johann Baptist Hofmann, and AntonSzantyr. 1963. LateinischeGrammatik, auf der Grund- lagedes Werkes von Friedrich Stolz und Joseph Hermann Schmalz: 1. BandLateinische Laut- und Formenlehre. Mu¨nchen:Beck’ sche Verlagsbuchhandlung. Levin,Beth, and Malka Rappaport. 1986. The formation of adjectival passives. LinguisticInquiry 17: 623–661. Levin,Beth, and Malka Rappaport-Hovav. 1995. Unaccusativity:At thesyntax –lexicalsemantics interface. Cambridge,Mass.: MIT Press. Lidz,Jeffrey. 1998. Valency in Kannada: Evidence for interpretive morphology. In UPennworking papers inlinguistics 5.2,37 –63.Penn Linguistics Club, University of Pennsylvania,Philadelphia. Lieber,Rochelle. 1980. The organization of thelexicon. Doctoral dissertation, MIT, Cambridge, Mass. Lindsay,William M. 1915. Ashorthistorical Latin grammar. Oxford:Oxford University Press. Marantz,Alec. 1984. Onthe nature of grammaticalrelations. Cambridge,Mass.: MIT Press. Marantz,Alec. 1988. Clitics, Morphological Merger, and the mappingto phonologicalstructure. In Theoreti- calmorphology, ed.Michael Hammond and MichaelNoonan, 253 –270.San Diego, Calif.: Academic Press. Marantz,Alec. 1989. Projection vs. percolation in the syntax of synthetic compounds. In Selectedpapers fromthe Annual Spring Colloquium, 95–112.UNC LinguisticsCircle, University of North Carolina, ChapelHill. FEATURES,SYNTAX,ANDCATEGORIESINTHELATINPERFECT 229

Marantz,Alec. 1994. A latenote on Late Insertion. Ms., MIT, Cambridge, Mass. Marantz,Alec. 1995. ‘ ‘Cat’’ asa phrasalidiom: Consequences of LateInsertion in DistributedMorphology. Ms.,MIT, Cambridge, Mass. Marantz,Alec. 1997. No escape from syntax: Don’ t trymorphological analysis in theprivacy of your own lexicon. In Proceedingsof the 21st Penn Linguistics Colloquium, 201–225.UPenn Working Papers inLinguistics, University of Pennsylvania, Philadelphia. Matthews,Peter H. 1972. Inflectionalmorphology: A theoreticalstudy based on aspects of Latin verb conjugation. Cambridge:Cambridge University Press. McCartney,Eugene S. 1926.Makeshifts for the passive of deponentverbs in Latin. PhilologicalQuarterly 5:289–298. McGinnis,Martha. 1998. Morphological ‘ ‘anaphora’’ assyntacticmovement. Ms., University of Pennsylva- nia,Philadelphia. Meillet,Antoine. 1966. Esquissed’ unehistoire de lalanguelatine. Paris:Klincksieck. No¨lting,Theodor. 1859. U¨ berdas lateinische Deponens. Wismar. Noyer,Rolf. 1992. Features, positions, and affixes in autonomous morphological structure. Doctoral dissertation,MIT, Cambridge, Mass. Noyer,Rolf. 1997. Features,positions and affixes in autonomous morphological structure. New York: Garland. Noyer,Rolf. 1998. Impoverishment theory and morphosyntactic markedness. In Morphologyand itsrelation tosyntax and phonology, ed.Steven Lapointe, Diane Brentari, and Patrick Farrell, 264 –285.Stanford, Calif.:CSLI Publications.[Distributed by Cambridge University Press.] Oltra,Maria Isabel. 1999. On the notion of themevowel: A newapproach to Catalanverbal morphology. Master’s thesis,MIT, Cambridge, Mass. Pesetsky,David. 1995. Zerosyntax: Experiencers and cascades. Cambridge,Mass.: MIT Press. Pollock,Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. LinguisticInquiry 20:365–424. Sauerland,Uli. To appear. Scope reconstruction without reconstruction. In Proceedingsof the 17th West CoastConference on FormalLinguistics. Stanford,Calif.: CSLI Publications. Serbat,Guy. 1980. Le parfait de l’indicatif actif. In Le sensdu parfait de l’ indicatifactif en latin, ed. Guy Serbat,1– 33. Colloque de Morigny, Universite ´deParisIV Sorbonne. Sicking,C. M.J.,andPeter Stork. 1996. The synthetic perfect in classical Greek. In Two studiesin the semanticsof the verb in classical Greek, ed.C. M.J.Sickingand Peter Stork, 119 –298.Leiden: E. J. Brill. Sihler,Andrew L. 1995. New comparativegrammar of Greek and Latin. New York:Oxford University Press. Sobin,Nicholas. 1985. Case assignment in Ukrainian morphological passive constructions. LinguisticInquiry 16:649–662. Sommer,Ferdinand. 1914. Handbuchder lateinischen Laut- und Formenlehre. Heidelberg:Carl Winters Universita¨tsbuchhandlung. Varro,Marcus Terentius. 1938. Onthe Latin language, with an English translation by Roland G. Kent. Cambridge,Mass.: Harvard University Press. Wackernagel,Jakob. 1904. Studien zum griechischen Perfektum. In KleineSchriften Band II. Go¨ttingen: Vandenhoeckand Rupprecht. Wackernagel,Jakob. 1920. Vorlesungenu¨ berSyntax, mit besonderer Beru ¨cksichtigungvon Griechisch, Lateinisch,und Deutsch. Basel:Philologischen Seminar der Universita ¨t Basel. Wasow,Thomas. 1977. Transformations and the lexicon. In Formalsyntax, ed.Peter Culicover, Thomas Wasow,and Adrian Akmajian, 327 –360.New York:Academic Press. 230 DAVIDEMBICK

Abbreviationsof ClassicalSources Cic. 4 Cicero L. 4 Livy Prisc. 4 Priscian Q. 4 Quintilian Sall. 4 Sallust

Departmentof Linguistics and Philosophy E39-245 77Massachusetts Avenue MIT Cambridge,Massachusetts 02139 [email protected]