Decomposing Pronouns Rose-Marie De´chaine MartinaWiltschko

Startingwith the idea that the notion ‘ ‘pronoun’’ isnot a primitiveof linguistictheory, we propose that it isnecessary to recognize (at least) threepronoun types: pro-DP, pro- fP,andpro-NP. Evidence support- ingthis three-way split comes from the sensitivity of certainproforms tothe predicate/ argumentdistinction, the internal structure of pro- forms,and the binding-theoretic properties of proforms.Recognizing differentpronoun types also sheds light on the formal (dis)similarities betweenobviation and switch-reference.

Keywords: pronoun,clitic, agreement, DP syntax,predicate/ argument asymmetries,binding theory, obviation, switch-reference

1TheProposal: Pro-NP, Pro- f P, Pro-DP We proposethat the notion ‘ ‘pronoun’’ isnot a primitive.Rather, pronoun types are defined morphosyntactically,and in turn, the morphosyntactic status of agivenpronoun type determines itsbinding properties. This analysis provides a principledsolution to the problem arising in theoriesthat, following Postal (1966) and Abney (1987), uniformly treat pronouns as DPs. Such theoriesclaim that differences in the distribution of pronounscorrelate with differences in their internalstructure (e.g., Cardinaletti 1994, Ritter 1995, Noguchi 1997). However, this begs the questionof howthe syntax can ‘ ‘see’’ theinternal structure of aDP. Inother words, attributing internalstructural differences to pronouns does not solve the problem of externaldifferences. Our solutionis to recognize the distinct categorical status of (atleast) three different pronoun types. Giventhe general proposal that ‘ ‘pronoun’’ isnot a primitive,we expectthat it is not a uniformsyntactic object. We arguethat this is correct: languages can have three pronoun types—pro-DP, pro- fP,andpro-NP— and each one is associated with a distinctsyntactic projec- tion,as in (1). Some pronouns have a trueDP shelland therefore function like ordinary R- expressions;these are pro-DPs, asin(1a). In addition, each subconstituent of the DP canfunction asa proformin its own right. These are pro- fPs andpro-NPs, asin(1b)and (1c). 1

Thanksto Eleanor Blain, Strang Burton, Re ´jean Canac-Marquis,Guy Carden, Anna Cardinaletti, Henry Davis, Anna Maria Di Sciullo,Jila Ghomeshi,Ken Hale, VictorManfredi, John Nichols, Tohru Noguchi, Yves Roberge,Johan Rooryck, SatoshiTomioka, Mireille Tremblay, Anne Zribi-Hertz, and audiences at theUniversity of Antwerp (NP-DP Conference), theUniversity of BritishColumbia (Department ofLinguisticsWednesday Research Seminar),the University of Manitoba (Department ofLinguistics Colloquium), and the Canadian Linguistic Association Conference (University of Alberta, Edmonton).This research was supportedby SSHRC MCRI grant 412-97-0016 (Principal Investigator: Anna Maria Di Sciullo,Universite ´ du Que´bec a‘ Montre´al),by UBC-HSS grant 5-71259 (Principal Investigator: Rose-Marie De´chaine, Universityof British Columbia), and by the Austrian Academy ofScience (APART435). 1 Fornow we abstract away fromthe distinction among phrasal pronouns, clitics, and agreement. As willemerge, the D/f/Ndistinctionthat we are postulatingcuts across these pronominaltypes (see sections4 and6).

Linguistic Inquiry, Volume 33, Number 3,Summer 2002 409–442 q 2002 bythe Massachusetts Institute ofTechnology 409

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 410 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(1) a. DP b. f P c. NP

D f P f NP N

f NP N

N

Thecategorical status of these pronominal categories determines their external syntax and theirinherent semantics, which in turn determines their binding-theoretic status. This is summa- rizedin (2). (2) Nominalproform typology Pro-DP Pro-fP Pro-NP Internalsyntax Dsyntax;morphologically neither D syntaxnor N Nsyntax complex syntax Distribution argument argumentor predicate predicate Semantics definite — constant Binding-theoreticR-expression variable — status

Apro-DP ispredictedto havethe syntax of adeterminer(phrase). Also, we claimthat pro- DPs willalways contain fPandNP assubconstituents. Given their external category as DPs, we furtherpredict that they will be restricted to argument position, on the assumption that DPs canonly be arguments (cf. Stowell1989, Longobardi 1994). 2 As for theirsemantics, DPs are demonstrablydefinite and consequently function as R-expressions, and so for thepurpose of bindingtheory they are subject to Condition C. We consider pro-fP tobe a coverterm for anyintermediate functional projection that intervenesbetween N andD andthat encodes f-features(where f-featuresinclude number and gender,and in somecases person). Pro- fPs arepredicted to have neither the syntax of determiners northat of nouns. We furtherclaim that there is no inherent restriction on their distribution; consequently,they can function either as predicates or as arguments. We arguethat they lack

2 We adopta strongversion of thesyntax/ semantics mappinghypothesis (see De´chaine1993). Note that both Stowell (1989)and Longobardi (1994) actually make aweaker claim, tothe effect thata DPcan beanargument, and that an NP cannot.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 411

inherentsemantics; that is, they simply spell out f-features.Their binding-theoretic status is that ofavariable,and so they correspond to the standard ‘ ‘ConditionB pronouns.’’ Finally,pro-NPs havethe same syntax as lexicalnouns, and as NPs we predictthat they will occurin predicate position (just as other lexical categories do). Syntactically, they are predicates; semantically,they are constants. We arguethat they are undefined with respect to bindingtheory; rather,their binding properties follow from theirinherent semantics in a predictableway. Inthe remainder of the article we presentevidence for theproposal that there are (at least) threepronoun types: pro-DP, pro- fP,andpro-NP. Onthe basis of evidencefrom differentlan- guages,we arguein section 2 thatthe pro-DP/ fP/NPdistinctionis necessary in order to account for thebehavior of different pronoun types. Having established the plausibility of the D/ f/N distinction,we thenexplore some consequences of theanalysis. In section 3 we showhow English pronounsare to be analyzedwithin the present proposal. In section 4 we considerhow the D/ f/ Ndistinctionsheds light on the Romance pronominal system (using French as theexample lan- guage).In section 5 we showhow recognizing different pronoun types allows us to formalize thesimilarities and differences between reference-tracking systems, in particular obviation and switch-reference.In section 6 we presentthe general conclusion and consider the broader implica- tionsof the analysis. Although we concentrateon the distinction among DPs, fPs, andNPs as itpertains to proforms,we suggestthat this tripartite division is observable not just with pronouns, butalso with other nominal expressions.

2Onthe Necessityof the Pro-DP/ f P/NPDistinction We arguethat the different behavior of pronouns in different languages reflects their categorical status.We first discussHalkomelem pro-DP, thenShuswap pro- fP,andfinally Japanese pro- NP.

2.1Pro-DPs: HalkomelemIndependent Pronouns Halkomelemis a CentralCoast Salish language; the data used are from theUpriver dialect (Sto´:lo¯Halq’eme ´ylem).It is a head-markinglanguage and is consistently predicate initial. In additionto pronominal clitics and affixes, Halkomelem has a setof independent(emphatic) pro- nouns,which are relevant for thepresent discussion. The Halkomelem independent pronoun paradigmis given in (3). 3

3 Abbreviationsused: 1 4 1stperson, 2 4 2ndperson, 3 4 3rd person, ACC 4 accusative, AGR 4 agreement, ARG 4 argument, COMP 4 complementizer, CONJ 4 conjunction, COP 4 copula, DEIC 4 deictic, DET 4 determiner, DS 4 different-subjectmarker, EMPH 4 emphatic, EXCL 4 exclusive, FEM 4 feminine, FOC 4 focus, FUT 4 future, GEN 4 genitive, LINK 4 linker, MASC 4 masculine, NEG 4 negation, NOM 4 nominative, NOML 4 nominalizer, OBJ 4 object, OBL 4 oblique, OBV 4 obviative, OP 4 operator, PAST 4 past tense, PL 4 plural, POSS 4 possessor, PRED 4 predicate, PRES 4 present, PROX 4 proximate, REDUP 4 reduplicative, SG 4 singular, SS 4 same-subject marker, SUBJ 4 subject, TNS 4 tense, TRANS 4 transitivizer.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 412 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(3) Halkomelemindependent pronouns (adaptedfrom Galloway1993:171 –172) Singular Plural

1 te-’e´lthe (DET-1SG) te-lhlõ´melh (DET-1PL) te-a´’elthe (DET-1SG,EMPH) 2 te-le´we (DET-2SG) te-lhwe´lep (DET-2PL) 3 tu´-tl’o‘ (DET-3SG) tu-tl’o´; lem (DET-3PL) thu´-tl’o‘ (DET.FEM-3SG) thu-tl’o´:lem (DET.FEM-3PL) yu-tl’o´:lem (DET.PL-3PL)

Halkomelemindependent pronouns show all the properties of fullDPs. First,they have D syntax andare morphologically complex. Second, they are restricted to argument position. Third, they havethe binding-theoretic status of R-expressions. 4 Thiscluster of propertiesis explainedby the hypothesisthat they are pro-DPs. 5 Onindependent grounds, Wiltschko (1998b, 2002) argues that Halkomelem independent pronounsare DPs withthe structure shown in (4).

(4) Pro-DP structure DP

D f P

tœ f NP

tl’ò À Thisstructure implies that independent pronouns are morphosyntactically complex. 6 The pronoun ismade up of a determiner—here tu´ —whichis syntactically visible. The rest of the pronoun (tl’o‘ )isidentified as pro- fPandspecifies person and number features (here 3rd singular). Thestructure in (4) furtherpredicts that independent pronouns contain an NP position.This NPpositioncan be either overt or covert, the latter resulting in the ‘ ‘pronominal’’ useof the pronoun.

(5) Tl’o´-cha-l-suqwemcõ ´we-t [thu´-tl’o‘ q’ami ]A R G . then-FUT-1SG-so hug-TRANSDET.FEM -3SG girl ‘ThenI’ mgoingto hug that girl.’ (Galloway 1993:174)

4 Definiteness,in the sense offamiliarity, is oftenlocalized in DP syntaxin many languages (e.g., English). As discussedat lengthin Matthewson 1998, (in)definiteness contrasts are notrelevant for Salish. 5 Thepro-DP analysis also extends to German d-pronouns(Wiltschko 1998a) and perhaps to Turkish (Baggaley 1998). 6 See Wiltschko1998b, 2002, for additional evidence supporting this proposal.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 413

The[Det N]sequencein (5) establishestwo things: first, that the Halkomelem independent pronoun thu´-tl’o‘ canfunction as an article; second, that the tl’o‘ subconstituentcannot be equated with a noun. TheDP analysisof Halkomelem independent pronouns straightforwardly explains why they havethe same syntactic distribution as full DPs. Giventheir categorical status as DP, we predict thatindependent pronouns are restricted to argument position. This prediction is borneout.

(6) a. [La´m]P R E D [tu´-tl’o‘ ]A R G . go DET-3SG ‘He goes.’(Galloway 1993:173)

b. *[Tu´-tl’o‘ ]P R E D -chate Bill kw’ e may-th-o´me. DET-3SG-FUT DET Bill COMP help-TRANS-2SG.OBJ Independentpronouns occur only in argument position, (6a). In (6b) the independent pronoun appearsin predicateposition and the result is ungrammatical.Note, however, that ‘ ‘pronominal’’ forms arenot excluded from predicateposition per se. 7 Rather,a subconstituentof theindependent pronouncan appear in predicate position —namely, tl’o‘ (7a),which we analyzeas a fP. Since fPs canfunction as arguments or predicates, we alsopredict that tl’o‘ canappear in argument position.However, this is not so, as shown in (7b).

(7) a. [Tl’o‘ ]P R E D -chate Bill kw’ e may-th-o´me. 3SG-FUT DET Bill COMP help-TRANS-2SG.OBJ ‘Itwill be Bill that helps you.’ (Galloway 1993:172)

b. *[La´m]P R E D [tl’o‘ ]A R G . go 3SG ‘He goes.’ ThatHalkomelem fPs arerestricted to predicate position is not the effect of anyinherent property of pro-fPs inthislanguage. Rather, we arguethat it reflects a generalmarkedness principle that governsblocking (Wunderlich 1996, Williams 1997). For concreteness,we adoptKoster’ s (1997) Principleof MaximalSpecialization. (8) Principleof Maximal Specialization Ina grammaticaldependency relation R, selectthe most specialized form. Aform A ismore specialized than B ifAcanfulfill fewer functionsthan B. (Koster 1997:224) Halkomelemhas both pro-DPs andpro- fPs. Giventhat pro-DP isspecializedfor theargument function,it follows that Halkomelem pro- fPisblocked from thisposition. Thismuch accounts for thedistribution of Halkomelemindependent pronouns: as pro-DPs,

7 Language-internalevidence supports the claim that tl’o‘ occupiesa predicate position.First, Halkomelem (likeall Salishlanguages) is strictlypredicate initial.Second, the future marker cha attaches topredicates. Third, if tl’o‘ is not analyzedas apredicate,then the example in(7a)will have no predicate inthe matrix clause.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 414 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

theymay function as articles, and they are restricted to argument position. Now considertheir bindingproperties. We proposedabove that R-expressions, as arguments, are to be defined as DPs. Consequently,we predictthat Halkomelem independent pronouns, as DPs, shouldbe subject toCondition C. Thisprediction is borneout.

(9) *Su´q’-t-es [te swõ´yeqe]i te kopu´-s [tu´-tl’o‘ ]i . search-TRANS-3.SUBJ DET man DET coat-3.POSS DET-3SG

?‘The mani was lookingfor his i coat.’(Wiltschko 1998a:444) Bythe same reasoning we furtherpredict that independent pronouns (as R-expressions)cannot functionas bound variables. This prediction is also borne out.

(10) *[Me´kw’ye swõ ´yeqe]i kw’a´kw’ets-et-es te sto ´les-s [tu´-tl’o‘ lem ]i . every DET.PL manlooking- TRANS-3.SUBJ DET wife-3.POSS DET-3.PL

?‘All meni arelooking at their i wives.’(Wiltschko 1998b:445) Havingshown that Halkomelem independent pronouns have all the properties of DPs, we now turnto Shuswap independent pronouns.

2.2 Pro-fPs: ShuswapIndependent Pronouns Shuswap(Secwepepemctsõ ´n)belongs to the Northern Interior branch of Salish, spoken in the interiorof British Columbia. Like Halkomelem, Shuswap is a head-markinglanguage (full DP argumentsare optional) and it is predicate initial. Arguments are marked on the verb as clitics oragreement affixes. In addition, there is a setof independent (emphatic) pronouns. They are listedin (11). (11) Shuswapindependent pronouns (adaptedfrom Kuipers1974, Lai 1998) Singular Plural

1n-tsets-we7(1 SG-EMPH-DEIC )wll-enwi7-kt( PL-EMPH-1PL) wll-enwi7-s-kucw( PL-EMPH-3-2EXCL) 2 7-enwi7 (2SG-EMPH) wll-enwi7-mp( PL-EMPH-2PL) 3 newi7-s (EMPH-3) wll-enwi7-s( PL-EMPH-3)

Shuswapindependent pronouns differ strikingly from theirHalkomelem counterparts. We suggestthat this is a reflexof a categoricaldifference. In particular, we proposethat Shuswap independentpronouns are of category fP,asin (12). 8

8 AlthoughShuswap independent pronouns are morphologicallycomplex, Lai (1998)argues that they are syntactic atoms.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 415

(12) Pro- P structure f P

f NP

ntsetswe7 À As pro-fPs, Shuswapindependent pronouns show the following cluster of properties. First, they haveneither D syntaxnor N syntax.Second, they can be predicates or arguments. Third, they actlike Condition B pronounsin thatthey can be bound outside their local domain, and they can functionas boundvariables. Evidencefor thisanalysis comes from thefollowing considerations. First, as Lai (1998) shows,independent pronouns do not have NP syntax,as can be seen by comparing them with trueNPs. Davis,Lai, and Matthewson (1997) argue that complex nominal predicates like the one in(13) must be of categoryN.

(13) Yirõ´7 te [sqe´lemcw]N l wõ´.w.k-t-sem-s. DEIC OBL man COMP see(REDUP)-TRANS-1SG.OBJ-3SG.SUBJ ‘That’s theman that saw me.’(Lai 1998:41, (39a)) Crucially,as Lai (1998) observes, Shuswap independent pronouns cannot appear in thisposition. (14) *Yirõ´7 te [newõ´7-s] wõ´.w.k-t-sem-s. DEICOBLEMPH -3 see(REDUP)-TRANS-1SG.OBJ-3SG.SUBJ ‘That’s HIM thatsaw me.’(Lai 1998:41, (39b)) Theill-formedness of (14) establishes that Shuswap independent pronouns are not of category N. Furthermore,Lai (1998) also shows that Shuswap independent pronouns do not have DP syntax.Evidence comes from thefact that they can be preceded by the same determiner as full NPs—namely, re.

(15) a. [Wõ´.w.k-t-À-en]P R E D [re n-tse´ts-we7]A R G . see (REDUP)-TRANS-3SG.OBJ-1SG.SUBJ DET 1SG-EMPH-DEIC ‘Isaw him.’(Lai 1998:28, (10))

b. [Wõ´k-t-À-s]P R E D [re John]A R G . see-TRANS-3SG.OBJ-3SG.SUBJ DET John ‘S/hesaw John.’(Lai 1998:11, (15)) If Shuswapindependent pronouns were themselvesDPs, we wouldnot expectthem to bepreceded bya determiner,as in (15a). And if theywere NPs, we wouldnot expect them to be blocked from nominalpositions, as in (14). Consequently, Shuswap independent pronouns are neither

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 416 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

DPs norNPs; we concludethat they are to be analyzed as fPs. Thisanalysis is also consistent withtheir external distribution. Shuswapindependent pronouns can function as predicates (a positionfrom whichDPs are excluded,as we saw inHalkomelem). 9

(16) [Newõ´7-s]P R E D [re wõ´k-t-À-m-es]A R G . EMPH-3 DET see-TRANS-3SG.OBJ-PAST -3SG.CONJ ‘Itis HIM thatsaw him/her.’(Lai 1998:28, (13a)) Furthermore,Shuswap independent pronouns can also function as arguments.

(17) a. [Wõ´.w.k-t-À-en]P R E D [newõ´7-s]A R G . see(REDUP)-TRANS-3SG.OBJ-1SG.SUBJEMPH -3 ‘Isaw HIM.’(Lai 1998:60, (74a))

b. [Newõ´7-s]A R G [wik-t-À-s]P R E D [re Mary]A R G . EMPH-3 see-TRANS-3SG.OBJ-3SG.SUBJ DET Mary ‘HE saw Mary.’(Lai 1998:28, (11c)) Thiskind of syntacticambivalence is acrucialproperty of fPs:they can function both as predicates andas arguments. Finally,the binding-theoretic properties of Shuswapindependent pronouns crucially differ from thoseof their Halkomelem counterparts, in a waythat is predictable from theircategory. First,Shuswap independent pronouns act like Condition B pronouns,in that they can be bound outsidetheir local domain.

(18) Tsut-Ài m qwetse´ts-Ài [newõ´7-s]i . say-3SG.SUBJPAST leave-3SG.SUBJEMPH -3

‘Hei said that HEi left.’(Lai 1998:31) Second,unlike their Halkomelem counterparts, Shuswap independent pronouns can function as boundvariables.

(19) [Xwexwe´yt]i re swetxwis-t- À-e´s [newõ´7-s]i re qe´7tse-si. all DET who like-TRANS-3SG.OBJ-3SG.SUBJEMPH -3 DET father-3.POSS

‘Everyonei likes HISi father.’(Lai 1998:32, (21b)) We proposethat the differences between Halkomelem and Shuswap independent pronouns follow from theirdifferent categories. Halkomelem independent pronouns are pro-DPs, whereasShuswap independentpronouns are pro- fPs. Therefore,the syntax of the latter is not reducible to DorN

9 Areviewer suggeststhat the predicate/ argumentdistinction might reduce toa focus/nonfocuscontrast. However, thisis misleading.Shuswap independent pronouns— whether they appear inpredicate orin argumentposition— translate intoEnglish as focusedconstituents. This reflects thefact thatShuswap is ahead-markinglanguage with pronominal agreement; thus,independent pronouns yield a contrastivereading. The Salish literature often describes these pronouns as emphatic.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 417

syntax:they may function as argumentsor aspredicates,and in binding-theoreticterms they are variables.

2.3Pro-NP: Japanese Kare Finally,we turnto the last type of proform, pro-NP. We arguethat Japanese kare instantiates thistype and that it has the structure in (20). (20) Pro-NPstructure NP | N | kare Pro-NPs arepredicted to havethe syntax of nouns.This is indeed the case. Kare canbe preceded byanadjective,a possessive,or ademonstrativepronoun (Kuroda 1965:105, Noguchi 1997:777). (21)a. tiisai kare small he ‘hewho is small’ b.watasi-no kare I-GEN he ‘myboyfriend’ c. kono kare this he ‘thisguy here’ As for itsbinding properties, recall that a pro-NP isundefined with respect to binding theory. Rather,its properties follow from itsinherent semantics. With this in mind, let us consider the boundvariable properties of kare. Becausethey are nouns, pro-NPs areinherently constants. Constants cannot function as boundvariables; as aresult, kare (andits corresponding feminine form kanozyo)cannotfunction asa boundvariable.

(22) a. *Daremoi-ga karei-nohahaoya-o aisite-iru. everyone-NOM he-GEN mother-ACC love-PRES ‘Everyoneloves his mother.’ (Noguchi 1997:770, (1a)) ? ; x,x lovesx’ s mother

b. *Dono zyoseii-mo [kanozyoi-gatensai-da to] omotte-iru. everywoman-also she- NOM genius-COP COMP think-PRES ‘Everywoman thinks that she is agenius.’(Noguchi 1997:770, (1b)) ? ; x,woman(x), x thinksthat x isagenius

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 418 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

Now considerthe coreference properties of kare. Kare hasreferential properties in that it connotesthe features [ MALE] and [MARRIAGEABLEAGE ](Sugamoto1989:270, as citedin Baggaley 1998:49).1 0 Referentialproperties are a prerequisitefor coreference.We arguethat this is precisely thereason that kare cansupport coreference and consequently does not show Condition C effects.

(23) a. Johni-ga karei-nohahaoya-o aisite-iru. John-NOM he-GEN mother-ACC love-PRES

‘Johni loves hisi mother.’(Noguchi 1997:770, (2a))

b. Maryi-ga [kanozyoi-gatensai-da to] omotte-iru. Mary-NOM she-NOM genius-COP COMP think-PRES

‘Maryi thinksthat she i isa genius.’(Noguchi 1997:770, (2b)) Previousanalyses of kare haveaccounted for itsexceptional behavior by stipulatingthat it must be A¯ free (Aounand Hornstein 1992:5), that it must be operatorfree (Katada1991), or that itmustbe discoursebound (Sportiche 1986). In a morerecent proposal Noguchi (1997) tries to reducethe binding properties of kare toitscategorical status as anN (as opposedto D). Noguchi proposesthat kare betreated as an N containedwithin a DP. As aresult,his analysis faces the problemdiscussed above: namely, if all pronouns have the structure of a DP, thenthe syntax cannottreat proforms such as kare, whichhave a vacuousDP layer,any differently than pronouns thathave a nonvacuousDP layer. Our pro-NP analysisof Japanese kare makespossible a moreprincipled account of its properties:it hasN syntax,and as a lexicalcategory it hasthe syntaxof apredicate.As asemantic constant,it cannot function as a boundvariable, accounting for itsapparent ‘ ‘operator-free’’ status.And because kare hasresidual semantic content, it can support coreference, accounting for itsdiscourse-bound property.

2.4 Summary Beforewe illustratehow this analysis extends to the pronominal inventory of English and French, letus summarize the results so far. We haveargued that different pronoun types are best analyzed asbelonging to different syntactic categories: pro-DP, pro- fP,andpro-NP. We havefurther arguedthat these different categories correspond to differences in the internal as well as the externalsyntax of proforms.In addition, we claimthat binding theory is sensitiveto categories,in thatR-expressions (nominal expressions subject to ConditionC) aredefined as DPs andvariables (nominalexpressions that are subject to ConditionB) aredefined as fPs. NPs areundefined with respectto binding theory; their behavior is determined by their inherent semantics as constants. Thisis summarizedin (24).

10 Gendermay be,but isnotnecessarily, a f-feature. Forexample, in German das Ma¨dchen ‘thegirl’ is grammatically neuterbut nevertheless denotes a [ FEMALE]individual.Such examples establishthat lexical genderfeatures neednot coincidewith grammatical gender.We assume thatthe [ MALE]feature associated with kare islexical gender.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 419

(24) Examplesof proforms and their properties Halkomelemindependent Shuswap independent Japanese Example pronouns pronouns kare Proform pro-DP pro-fP pro-NP Internalsyntax Dsyntax;morphologically neither D syntaxnor N Nsyntax complex syntax Distribution argument argumentor predicate predicate Semantics definite — constant Binding-theoreticR-expression variable — status

As for thedistinction between predicates and arguments, we departfrom Longobardi’s (1994) claimthat all argument expressions are DPs. We claim,as Longobardi does, that a DP mustbe anargument (25a) and an NP mustbe apredicate(25b). However, we alsoclaim that fP is type flexible;it can be an argument or a predicate.Consequently, not all argument expressions are DPs (25c),and not all nominal predicates are NPs (25d). (25) a. DP Argument ! b. NP Predicate ! c. Argument DP, fP ! d.Nominal predicate NP, fP ! 3ThePronominal Inventory of English We nowturn to the pronominal inventory of Englishand consider how differentEnglish pronouns areto be analyzed within the present proposal. We arguethat in itsanaphoric use, one is a pro- NP.1 1 We furtherpropose that the English person pronoun inventory consists of pro-DPs (1stand 2ndperson pronouns) and pro- fPs (3rdperson pronouns).

3.1 English One as a Pro-NP FollowingPostal’ s (1966)proposal, we showthat English one hasthe syntax and semantics of atruepronoun and as such is to be analyzed as apro-NP. (26) NP-structure NP | N | one

11 The proform one is distinctfrom the numeral one, as illustratedin (i).

(i)the three large cars andthe [ NUM one] small [NP one] . . .

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 420 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

Evidencefor analyzing one asa pro-NP comesfrom thefollowing considerations. First, pro- NPs areexpected to havethe syntax of nouns.Consistent with this is thefact that one may follow adeterminer,a quantifier,or a modifier. (27) a. the one b. someone c. the real one Second,there is crosslinguistic evidence for treating one asa pro-NP. Incontexts where languagessuch as French allow a nullelliptical noun, English requires an overt proform in the form of one.

(28)a. La grande [ fille]i nepeut pas supporter la petite [ À]i. thetall girl NEG can NEG supportthe small b. La [voiture]i rougeest plus che ‘re que la [À]i jaune. thecar red is more expensive than the yellow

(29)a. The large [ girl]i can’t standthe small [ one]i . (Postal1966:202, (2))

b. The red [car]i ismoreexpensive than the yellow [ one]i . Letus now turn to the binding properties of one. Recallthat pro-NPs areundefined with respectto binding theory and that their binding properties follow from theirinherent semantics. Withthis in mind, consider the status of one withrespect to boundvariable anaphora. The inherent semanticsof a pro-NP isthatof a constant.Because constants cannot function as boundvariables, one (likeJapanese kare)cannotfunction as a boundvariable.

(30) a. *[Everybody]i thinks [one]i is a genius. ? ; x,xthinksthat x isagenius

b. *[Everybody]i loves [one]i’s mother. ? ; x,xlovesx’ s mother Eventhough as pro-NPs, English one andJapanese kare areboth defined as constants, these twoinstances of pro-NP differin their referential properties. As notedearlier, kare hasresidual referentialcontent and is therefore able to supportcoreference. In contrast, English one is a pure spell-outof N andso has no referential content. Since referential content is a prerequisitefor coreference,it followsthat one cannotsupport coreference and so cannot be boundby ananteced- ent.This results in apparent Condition C effects.

(31) a. *[Mary]i thinks [one]i is a genius.

b. *[Mary]i loves [one]i’s mother. Havingestablished that English one isanalyzableas apro-NP, we nowturn to theproperties of Englishpersonal pronouns, arguing that they instantiate pro-DP andpro- fP.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 421

3.2English Personal Pronouns: Pro-DPs andPro- fPs Englishpersonal pronouns are not a syntacticallyhomogeneous group (see, e.g., Ritter 1995). Oneexample of this heterogeneity is the fact that, in Standard American English, while plural 1stand 2nd person pronouns can function as determiners(i.e., precede nouns), 3rd person pronouns cannot(see Postal1966). 1 2 (32) a. we linguists us linguists StandardAmerican English b. you linguists you linguists (dialect A) c. *they linguists *them linguists We proposethat the contrast in (32) is a reflexof the different category (and thus structure) associatedwith each of theseproforms (see Ritter1995 for asimilarapproach). In particular, we arguethat 1st and 2nd person pronouns instantiate DPs and3rd person pronouns instantiate fPs, asshown in (33). (33) a. DP structure: 1st/2nd person b. P structure: 3rd person DP f P

D f P f NP

we f NP they À

N

À/linguists

We observethat while pro-DPs (1stand 2nd person pronouns) make an overt NP subconstituent available,pro- fPs do not. Insome varieties of American English, it is also possible for 3rdperson plural them (but not they)tocombine with N, asillustrated in (34) and (35). (34) a. we linguists us linguists dialect B b. you linguists you linguists c. *they linguists them linguists

12 We haveno account for the impossibility of combininga singularpronoun with a lexical noun:* I linguist, *you linguist, *himlinguist.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 422 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(35) a. [Them linguists]are subversive. b. [Them linguists],John likes ’ em. (GuyCarden, personal communication) We suggestthat the dialect B patternis consistent with the present proposal, inasmuch as them canbe decomposed into a boundD-morpheme th- and a clitic f-morpheme ’em. Independent evidencefor thisdecomposition comes from thefact that, in English,both the 3rd person singular andthe 3rd person plural pronouns have phonologically reduced clitic forms (Selkirk1984), as illustratedin (36). In our analysis these instantiate pro- f.

(36)a. I like[ f ’im] [ayla´ykIm]

b. I like [f ’em] [ayla´yk@m] Thedecomposition of them into th-em isfurther supported by the fact that th- isparadigmatic. Inaddition to occurring with the definite determiner, (37a), it occursthroughout the demonstrative series,(37b). We concludethat dialect B hasgeneralized the [D- f]structureto the pronoun them, (37c).An automatic consequence of thisreanalysis is that this dialect permits them to combine withovert nouns (e.g., themlinguists ).

1 3 (37) a. [D th-e] dialect B

b. [D th-[ f is]] [D th-[f ese]]

[D th-[f at]] [D th-[f ose]]

c. [D th-[f em]] Thatdialect B stillexcludes * theylinguists isconsistent with the fact that this pronoun does not havea transparent f-constituent,as confirmedby theill-formedness of * ’ey asa reducedpronoun. Inboth dialects 1st/ 2ndperson pronouns are able to combine with overt nouns, consistent withtheir pro-DP status.And in both dialects the 3rd person series shows a differentbehavior. Indialect A both3rd person proforms fail to cooccur with an overt noun, consistent with their pro-fPstatus.And in dialect B nominative they actslike a pro- fPinnot combining with an overtnoun, but accusative them actslike a pro-DP incombining with an overt noun. That the proformsshould be heterogeneous in this way —withnot only 1st/ 2ndperson contrasting with 3rdperson, but also different 3rd persons contrasting with each other —isexpected in a theory whereproforms are defined categorically. In principle, nothing prevents a languagefrom having apro-DP anda pro- fPinthesame ‘ ‘person,’’ withcorresponding syntactic and semantic differ- ences.Indeed, as we haveshown, this is true of Halkomelem, and it is also true of languages withobviation and switch-reference (see section5). Givenour claim that different English pronouns are associated with different structures, we expectthat these structural differences will correlate with binding-theoretic properties. In the

13 Areviewer suggeststhat the couldbe analyzedas th- ` he. Asimilar treatment is proposedby Kayne (2000) forthe French argument clitics le and la, whichhe decomposes as l-e and l-a, withthe vocalic elements encodingmasculine andfeminine gender, respectively. Transposing this analysis to English,we wouldthen expect a contrastbetween th- ` he and th- ` she, contraryto fact. Instead,we treat theschwa ofthe[ D@]as aPFspell-out.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 423

remainderof thediscussion, we concentrateon the difference between 1st/ 2ndperson pronouns (as pro-DPs) and3rd person pronouns (as pro- fPs), asthiscontrast appears in bothdialects. We startby considering the binding properties of 3rdperson pronouns. 3.2.1English 3rd Person Pronouns Are Pro- fPs Since fPs aredefined as variables, we predict that3rd person pronouns (as pro- fPs) willparticipate in bound variable anaphora, which is indeed the case.

(38) [Everycandidate ]i thinksthat [ he]i will win. ; x,candidate(x), x thinksthat x willwin Inaddition, 3rd person pronouns support coreference, in that they may be bound outside their localdomain.

(39) [John]i thinksthat [ he]i will win. Itis the intersection of these two properties— the ability to function as a boundvariable and the abilityto support coreference —thatcorresponds to the classical Condition B pronoun. 3.2.2English 1st /2ndPerson Pronouns Are Pro-DPs As discussedabove, the distributional evidencesuggests that 1st and 2nd person pronouns are pro-DPs. If so,and if DPs aredefined asR-expressions, this makes two predictions. First, 1st/ 2ndperson pronouns should not be able tofunction as bound variables. Second, 1st/ 2ndperson pronouns should be subjectto Condition C. Thefirst predictionis borneout. Consider the following sentence and its potential interpreta- tions:

(40) Ii knowthat John saw mei ,andMary does too. 4 a.‘ Iknowthat John saw me,and Mary knows that John saw me.’ lx[xknows that John saw me] & ly[yknows that John saw me] ? b.‘ Iknowthat John saw me,and Mary knows that John saw her.’ lx[xknows that John saw x]& ly[yknows that John saw y] Thesentence in (40) involving VP-ellipsis can receive a strictidentity reading (a), butnot a sloppyidentity reading (b). This indicates that the 1st person pronoun ( me)cannotbe construed asa boundvariable, as predictedby the present analysis. Turningto thesecond prediction —namely,that 1st/ 2ndperson pronouns will be subjectto ConditionC —we immediatelyface a problem.The following examples show that 1st and 2nd personpronouns can be nonlocallybound.

(41) a. Ii saidthat John saw mei .

b. Youi saidthat John saw youi . Butif 1stand 2nd person pronouns are DPs, andif DPs aredefined as R-expressions, then why dothey differ from names?In particular, the latter cannot be nonlocally bound.

(42) *Shei saidthat John saw Maryi .

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 424 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

If we adopta standardbinding theory, where R-expressions are subject to ConditionC, then1st and2nd person pronouns (as R-expressions)should be ungrammaticalin sentences like (41a –b), contraryto fact. 1 4 Inthe next section we showthat this problem is not an artifact of thepresent analysis,and we sketcha possiblesolution that is compatiblewith our general proposal. 3.2.3The Condition C Problemwith ‘ ‘FullDPs’ ’ Theproblem we havenoted for English1st and2nd person pronouns has also been identified for fullDPs byDemirdache(1997a). She shows thatfull DPs inLillooet, a NorthernInterior Salish language, have the following properties. First, theycannot function as bound variables.

(43) [Swat]i kuzwa ´t-en-[ as]* i/j kw-s xwey-s-[ as]i k-Wany? who LINK know-TRANS-3SG.SUBJDET-NOML love-TRANS-3SG.SUBJ DET-Wany

a. *‘Whoi does hei know ti lovesWany?’

b. ‘Whoi does hej know ti lovesWany?’ (adaptedfrom Demirdache1997a:60, (15)) (43)shows a standardstrong crossover (SCO) violationand thus establishes that full DPs cannot functionas bound variables. Second, Lillooet DPs donot seem to obey Condition C: theycan benonlocally bound.

(44) Sqwa´l’-en-[as]i s-Buckykw-s nilhs-[ Wany]i ta qwatsa´ts-a. say-TRANS-3SG.SUBJ NOM-Bucky DET-NOMLFOCNOML -Wany DET leave-DET

‘Shei toldBucky that it’ s Wany i thatleft.’ (Demirdache 1997a:54, (6)) Demirdache’s (1997a)analysis of the Lillooet binding facts is as follows. She argues (follow- ingReinhart 1986) that grammar onlyregulates bound variable anaphora. In addition, she assumes thatEnglish DPs arequantificational and consequently undergo quantifier raising (QR). As a result,classical Condition C effectsare analyzed as SCO violations.

(45)a. *I knowhe i loves Oscari .

b.LF: *[Oscar] i [I know hei loves ti ] Incontrast to English DPs, LillooetDPs arenot quantificational (Matthewson 1998) and so do notundergo QR. Demirdacheargues that it isthe nonquantificational status of Lillooet DPs that isresponsiblefor theapparent Condition C violations:Lillooet DPs neverundergo QR, sothey neverinduce SCO violations. 3.2.4The Analysis of English1st /2ndPerson Pronouns As Demirdachepoints out, it is not the casethat all English DPs arequantificational. Rather, English focused DPs anddeictics, like LillooetDPs, donotexhibit SCO violations.The inherently deictic nature of 1stand 2nd person pronounsimplies that they too are not quantificational. As aconsequence,1st and 2nd person pro-DPs donot undergo QR andtherefore do not induce SCO violations.

14 Apragmaticallybased analysis such as Reinhart’s (1986)also fails toaccount for crosslinguistic differences between proformsthat do notsupportcoreference (e.g.,Halkomelem pro-DPsthat show Condition C effects) andproforms thatdo support coreference (e.g.,English 1st and 2nd person pro-DPs that do notshow standard Condition C effects).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 425

(46) a. Ii saidthat you saw me i .

b. LF: Ii saidthat you saw me i Thisimmediately accounts for thelack of ConditionC effectswith 1st and 2nd person pronouns andallows us to maintain the claim that they are pro-DPs. Demirdache’s proposalprovides a morenuanced picture of coreferenceeffects. As mentioned above,we followReinhart (1986) in takingbound variable anaphora to be the essence of binding theory.However, following Demirdache (1997a), we claimthat coreference is regulated in one oftwo ways. First, it may fall within the domain of bound variable anaphora, in which case disjointreference is aby-productof SCO violations.This only arises if therelevant DPs undergo QR,for example,English nonfocused nondeictic DPs. Second,if therelevant DPs donotundergo QR—for example,English focused DPs anddeictics, English person pro-DPs, andall Lillooet DPs—the coreference possibilities are determined by pragmatic coreference.

3.3On the Predicative Status of English Pro- fPs We nowturn to Englishpro- fPs. We claimthat pro- fPs canbe argumentsor predicates, while pro-DPs canonly function as arguments. This leads us toexpectthat English 3rd person pronouns, whichwe analyzeas pro- fPs, shouldoccur in a widerrange of contexts than pro-DPs. English3rd person pronouns can function as arguments.

(47) [He]A R G saw [her]A R G . Theability of pro- fPs tofunction as predicates is confirmedby the contrast in (48).While a 3rd personpronoun ( fP)canoccur in predicate position, a baredemonstrative (DP) cannot.

(48)a. That’ s [ her]P R E D .

b. *She’s [that]P R E D . We acknowledgethat (48) does not provide conclusive evidence for thepredicative status of English3rd person pronouns. In particular, it is alsopossible for 1st/2ndperson pronouns— which we analyzeas pro-DPs—tooccur in postcopular position. (49)a. That’ s [ me].

b. *I’m [that]P R E D . (50)a. That’ s [ you].

b. *You’re [that]P R E D . Tentatively,we suggestthat (49a) and (50a) are well formed as equative structures involving two DPexpressions. 1 5

15 TheEnglish data are confoundedby thefact thatEnglish intonational focus provides a mechanism foroverriding the‘ ‘normal’’ subject-predicateorder. In languages where thisstrategy is notavailable (e.g., Salish and French), the predicate statusof fPscan bemore clearly established.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 426 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

Afurtherargument for thepredicate status of English3rd person pronouns comes from the factthat such pronouns can participate in word formation. (51) a. [she]-male [she]-society [she]-oak b. [he]-goat a real [he]-man [him]-bo(vs. bimbo) c. The [hes]wouldquarrel and fight with the females. (JonathanSwift, cited in Webster’ s ThirdInternational Dictionary ) Suchexamples establish that pro- fPs canbe property denoting, consistent with our analysis.Note moreoverthat we neednot invoke any special mechanism (e.g., semantic type-shifting) to account for thepredicative nature of pro- fPs. Inthe present proposal 3rd person pronouns are analyzed as fPs and fPs aretype flexible: they can function as predicates or asarguments. Crucially,there are no examples involving English 1st/ 2ndperson pronouns as parts of nominalcompounds. (52) a. *[me]-male b. *[you]-goat Inour analysis the impossibility of (52a– b) follows from thepro-DP statusof English 1st/ 2nd personpronouns. 1 6 Summingup the results for English,we concludethat the syntax and semantics of the pronominalinventory support the D/ f/Npartition,with pro-DP correspondingto 1st/2ndperson pronouns,pro- fPto3rd person pronouns, and pro-NP to one.

4FrenchClitic Proforms We nowturn to the pronominal inventory of French. We haveso far abstractedaway from the distinctionamong phrasal pronouns, clitics, and agreement. As we willshow, the D/ f/Ndistinction thatwe arepostulating cuts across these pronominal types. In particular, Romance clitic pronouns showa divisionbetween pro-N andpro- f thatis predictedby ourtheory. Although the data are drawnfrom French,to ourknowledgemost of thepropertiesthat we discusshold across Romance. We firstshow that French en hasall the properties of a pro-N clitic.We thenestablish that all otherpronominal clitics are pro- f.

16 As pointedout by ananonymous reviewer, expressionssuch as ame-first attitude are apparentcounterexamples tothe claim that1st/ 2ndperson pronouns do not participate in compounding. We take suchexamples tobe phrasal compounds(see Di Sciulloand Williams 1987). As aresult,any XP, includingfunctional categories, may participatein these compoundstructures. (i)a. a[holier-than-thou]attitude b.an [I-don’ t-give-a-flying-fuck] attitude

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 427

4.1 En asaPro-NClitic Thatat least some occurrences of the clitic en correlatewith N syntaxis widely recognized in theliterature (Kayne 1975, Postal 1994, Pollock 1998). 1 7 For example, en maysubstitute for a lexicalnoun in the context of adjectival (weak) quantifiers,as well as modifying adjectives. 1 8 (53)a. J’ aivu trois/ unlivre(s). Ihaveseen three/ onebook(s) b. J’en aivu trois/ un. I en haveseen three/ one (54)a. J’ aivu plusieurs livres. Ihaveseen several books b. J’en aivu plusieurs. I en haveseen several

17 Pollock(1998) identifies three contextsfor clitic en:adverbial,adnominal, and quantitative. Examples of each are givenin (i)– (iii). In this article we discussonly quantitative en. (i)a. De ce fait,Jean aavale´ sonchapeau. ofthis fact Jean has swallowedhis hat (Pollock 1998:300, (1b)) b. Jean en aavale´sonchapeau. Jean en has swallowedhis hat (Pollock 1998:300, (1a)) (ii)a. J’ai lule premier chapitrede ce livre. Ihaveread thefirst chapter of this book b. J’en ai lule premier chapitre. I en haveread thefirst chapter (iii)a. J’ai lutrois livres. Ihaveread three books b. J’en ai lu trois. I en haveread three 18 Notethe following contrast between French en and English one: (i)a. J’ai achete´ unevoiture rouge, et Marie ena achete´ une jaune. Ihavebought a car redand Marie en has boughta yellow b.J’ ai achete´ la voiturerouge, et Marie aachete´ la jaune À. Ihavebought the car redand Marie has boughtthe yellow c. *J’ai achete´ la voiturerouge, et Marie ena achete´ la jaune. Ihavebought the car redand Marie en has boughtthe yellow (ii)a. Iboughta redcar, andMary bought a yellowone. b.I boughtthe red car, andMary bought the yellow one. Whereas English one can appear witheither a definiteor an indefinite phrase, French en can appear onlywith an indefinite. Inthe presence ofadefinitedeterminer, pro-N is null.We suggestthat this has todo with the restriction on moving out ofadefiniteDP (Diesing1992). This is consistentwith the fact that en cannotbe construed with strong quantifiers. (iii)a. *J’enai vutous. I en haveseen all b.*J’ enai vuchacun. I en haveseen each

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 428 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(55)a. J’ aivu un grand livre. Ihaveseen a largebook b. J’en aivu un grand. I en haveseen a large Consistentwith the claim that en isapro-N clitic,observe that, similar to English one, it cannot functionas a boundvariable, nor can it support coreference. These facts are illustrated in (56) and(57), respectively.

(56) a. *[Chacun]i penseque Jean [ en]i a vu. each.onethinks that Jean en has seen

b. *[Quelqu’un]i penseque Jean [ en]i a vu. someonethinks that Jean en has seen

(57) a. *[Marie]i penseque Jean [ en]i a vu. Mariethinks that Jean en has seen

b. *[Des e´tudiants]i pensentque Jean [ en]i a vu. somestudents think that Jean en has seen

4.2French Clitics as Pro- f Havingestablished that en isa pro-N,we nowconsider the other pronominal clitics, which we analyzeas pro- f.Inthis respect we againdepart from standardtreatments, which date back to Postal1966. Such proposals often invoke the parallel between clitics and articles as areasonto treatclitics as belonging to thecategory D. Althoughwe agreethat Romance clitics and articles arecategorically the same, we treatthem as belonging to the category f (ratherthan D). We arguethat this accounts for thedistribution of pronominal clitics on the one hand, and for thereferential defectiveness of Romance articles on the other hand (Vergnaud and Zubizarreta 1992). Thediscussion proceeds in two steps. We firstshow that 3rd person clitics ( l-clitics)have theattributes of pro- f,andwe thenextend the analysis to 1st/ 2ndperson clitics. 4.2.1 l-Clitics Inlight of the present proposal, it is significantthat French l-cliticscan be pro- argumentsor pro-predicates(De ´chaine1993). Their pro-argument status is illustratedin (58),and theirpro-predicate status in (59). (58)a. Jeanne la voit. Jeanneher sees b. Jeanne le voit. Jeannehim sees (59)a. Marie est une avocate, et Jeanne le/*la sera aussi. Marieis a lawyer( FEM)andJeanne it will.be too b.Jean est un avocat, et Franc¸ois le sera aussi. Jeanis a lawyer( MASC) and Franc¸oisit will.be too

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 429

A pro-f analysiscaptures the fact that l-cliticscan function as pro-arguments or aspro-predicates: argumentalpro- f inflectsfor gender;predicative pro- f does not.1 9 Consequently,there is no needto stipulate the existence of twohomophonous le clitics.2 0 The French l-cliticsconfirm our claim that fPs canfreely occur in predicate or argument position.This departs from analysesthat treat all pronominals as DPs, whichcan only be argument- typeexpressions. Alsoconsistent with the fPanalysisof l-cliticsis the fact that they can function as bound variables.

(60) a. [Chaquehomme ]i pense qu’[il]i estintelligent. eachman thinks that he is intelligent

b. [Chaquehomme ]i penseque Marie [ l]i’a vu. eachman thinks that Marie him has seen If we takeseriously the fact that the l-cliticshave the same form asarticlesin French, then we expectthat the latter will also be fPs. Thispredicts that French articles (as fPs) willdiffer inboth their syntax and their semantics from English-typearticles (as DPs). Thisis consistent withthe fact that Romance articles appear in a broaderrange of contexts than articles in other languagesdo (e.g., English, German). Vergnaudand Zubizarreta (1992) propose that there is a crosslinguisticdifference regarding whether a determinercan have an expletive use or not. For example,the French l-articledoes not have a fixedinterpretation. In some contexts it may be construedas adefinite,(61a); in other contexts it isambiguous between a genericand a definite construal,(61b). (61)a. Jean a achete´ le vin. Jeanhas bought DET wine ‘Jeanbought the wine.’ b. & Jeanaime le vin. Jean likes DET wine 4 i.‘ Jeanlikes wine.’ 4 ii.‘ Jeanlikes the wine.’

19 Thatthe pro-predicate clitic le doesnot inflect for gender might reflect theimplicational relation between Case andagreement: ifCase, thenagreement. If so,then a pro-argumentclitic shouldshow agreement (since itrequires Case), whereas apro-predicateclitic shouldnot. 20 Inour analysis the parallel between pro-predicateand pro-argument clitic isnot accidental: theyboth show nominal agreement, whichwe identifywith fP.Thus,the agreement morphologyassociated withnouns and adjectives has the same formal basis as theagreement morphologyassociated withclitics: all instantiate fP,as in(i). Accordingly, we adoptan analysis where NandA are bothdefined as [ `N]andmay bothbe contained in a fPprojection:[ f P . . . [+N . . . ]]. (i) l-es bell-es fill-es DET-PL beautiful-FEM.PL girl-FEM.PL ‘thebeautiful girls’

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 430 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

Itis the ambiguity of (61b) that distinguishes French articles from theirEnglish counterparts. Thisin turn reflects a categoricaldifference between English and French articles: the former belongto the category D, thelatter to the category f.As membersof f,Frencharticles do not havea fixedreferential value; this accounts for whythey are not inherently definite. 2 1 4.2.2 1st/2ndPerson Clitics We nowturn to French 1st/ 2ndperson clitics, arguing that they arealso pro- f.Oneof thestriking properties of Romance1st/ 2ndperson clitics is thatthe same forms canbe usedfor disjointreference, (62a), or for coreference,(62b). 2 2

(62)a. Marie i mj’a vu. b. Jei mei suis vu. Marieme has seen I me am seen ‘Mariesaw me.’ ‘Isaw myself.’

Mariei tj’a vu. Tui ti’es vu. you-SG you-SG

Mariei nousj a vu. Nousi nousi sommes vu. us us

Mariei vousj a vu. Vousi vousi eˆtes vu. you-PL you-PL We takethe possibility of local binding to reflect the pro- f statusof French 1st/ 2ndperson clitics.2 3 If theseFrench pronominals are pro- fs,then they should be ableto have a boundvariable interpretation,(63). 2 4

21 FollowingLongobardi (1994), we take thedefinite construal to reflect thepresence ofa nullD position,(i). When theD superstructureis absent,the generic reading becomes available,(ii).

(i) [D À [f le [NP vin]]] 4 ‘the wine’ (ii) [f le [NP vin]] 4 ‘wine’ 22 We donot take theselection of theauxiliary avoir in (62a) and eˆtre in(62b) to indicate transitivity and intransitivity, respectively.As evidencefor this, note that the question of whatdetermines auxiliaryselection is independentof binding. As iscommon in Romance, auxiliaryselection correlates withunaccusative versus unergative syntax, (i). However, this distinctionis not(entirely)lexically determined, given minimal pairssuch as (ii).(See Borer1994 for further discussion.) Inthe 3rd person, in addition to a difference inauxiliary selection, there is suppletion:the l-clitic marks disjointreference, while the s-clitic marks localbinding, (iii). Traditional descriptions often treat the s-clitic as inherentlyanaphoric, but thisis clearly anoversimplification, as itcan alsomark impersonalsubjects (e.g., Italian, Spanish) and 3rd person possessors(e.g., son livre ‘his/herbook’ ). (i)a. Ellea ri. b.Elle est arrive´e. she has smiled she is arrived ‘Shesmiled.’ ‘Shearrived.’ (ii)a. Ellea tombe´. b.Elle est tombe´e. she has fallen she is fallen ‘Shefell.’ (agentive) ‘Shefell.’ (nonagentive) (iii)a. Ellel’ a vu. b.Elle s’ est vue. she himhas seen she herself is seen ‘Shesaw him.’ ‘Shesaw herself.’ 23 If thepossibility of locally binding 1st/ 2ndperson pronouns is a diagnosticfor fP,thenthe following languages arguablyhave pro- fP:Haitian(De ´chaineand Manfredi 1994), Guadeloupe ´en (Bernabe´ 1983:918–920),Seychellois (Corne1977), Mauritian (Corne 1988), Niuean (Seiter 1980:78–79),Chamorro (Chung 1989:149), and German. 24 Notall speakers accept thebound variable construal. Similarly, Bouchard (1984) observes that 1st/ 2ndperson

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 431

& (63) [Je]i penseque la police [ m]i’avu,et Marie le pense aussi. Ithinkthat the police me have seen and Marie it thinks also 4 a.‘ Ithinkthat the police saw me,and Marie thinks that the police saw me.’ lx[xthinks that the police saw me] & ly[ythinks that the police saw me] 4 b.‘ Ithinkthat the police saw me,and Marie thinks that the police saw her.’ lx[xthinks that the police saw x]& ly[ythinks that the police saw y] Notecrucially that a boundvariable interpretation is unavailablefor English1st/ 2ndperson.

(64) [I]i thinkthat the police saw [ me]i ,andMary does, too. 4 a.‘ Ithinkthat the police saw me,and Mary thinks that the police saw metoo.’ ? b.‘ Ithinkthat the police saw me,and Mary thinks that the police saw hertoo.’ Inthe present analysis the impossibility of (64b) reflects the pro-DP statusof English 1st/ 2nd person. Anotherargument in favor of thepro- f statusof French1st/ 2ndperson clitics is thatthey mayparticipate in word formation, as in (65). Again, there is a contrastbetween English and Frenchin this regard. (65)a. tutoy-er ‘you-ing’(i.e., addressing someone with tu) vousvoy-er ‘pluralyou-ing’ b.tutoie-ment ‘you-ment’ vousvoie-ment ‘pluralyou-ment’ Thesecrosslinguistic differences establish that 1st/ 2ndperson pronouns are not inherently pro- Ds or pro-fs. Thelogic of the D/ f/Ndistinctionhas ledus toproposea novelanalysis of Frenchpronominal cliticsand articles. Rather than reducing clitics to their corresponding ‘ ‘determiners,’’ we have arguedthat clitics are pro- fs,as aretheir corresponding articles. For clitics,this has the advantage ofautomaticallyderiving their distributional and binding properties. 2 5 For articles,their f status correlateswith their defective referential properties.

clitics haveboth strict andsloppy identity interpretations in gapped VPs. (i)Je me trouvaisbe ˆte et ma soeuraussi. Ime foundstupid and my sister too 4 a. ‘Iconsideredmyself stupidand my sister consideredme stupid.’ lx[xconsider me stupid]& ly[yconsider me stupid] 4 b.‘ Iconsideredmyself stupidand my sister consideredherself stupid.’ lx[xconsider x stupid]& ly[yconsider y stupid] (adaptedfrom Bouchard 1984:60, (131)) 25 We haveno account for the suppletion of 3rdperson in the possessor paradigm, where son appears insteadof the l-based *lon, as discussedin Kayne 2000. While our proposal accounts for the possibility of local binding with 1st/ 2nd person (Je me vois ‘Isee myself’, Tu te vois ‘Yousee yourself’), Kayne’s proposaldoes not (Kayne 2000:162, fn. 82). Thetwo proposals converge on theconclusion that 1st/ 2ndperson clitics are notpro-Ds. However, Kayne treats l-clitics as pro-Ds,while we treat them as pro- fs.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 432 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

5Extendingthe Analysis:Obviation and Switch-Reference The D/f/Ndistinctionprovides a wayof understanding how properties that are usually viewed asunrelatedto each other— the definiteness of certainpronouns, the morphological complexity ofcertainpronouns, the distribution of certainpronouns, and the binding properties of different kindsof pronouns—can be captured in a nonstipulatoryfashion. We nowconsider two cases of referencetracking: obviation and switch-reference. Stirling (1993) observes that such devices are functionallysimilar inasmuch as they signal disjoint reference. In a languagethat exploits obvia- tionmarking, an obviative-marked argument is obligatorily disjoint from aproximate-marked argument,(66a). In a switch-referencesystem, if the subject of a dependentclause is distinct from thesubject of a mainclause, then this is obligatorily coded by different-subject marking, (66b). (66) a. Obviation

. . . , kaˆ-waˆpamaˆ-[t]i eˆh-kitoˆweˆhkwaˆmi-[yit]j . Plains Cree COMP-see-3.PROX COMP-snore-3.OBV

‘. . . , hei(PROX) saw that hej(OBV)was soundasleep and snoring.’ (Long1999:98, (17), from P:80-23) b. Switch-reference nya-isvar-miima-k Mojave when-sing-DS dance-TNS

‘When hei sang, hei danced.’ (Munro1980:145, (4); as cited in Stirling 1993:3, (3)) Stirlingnotes that any attempt to reduce these reference-tracking systems to a singlecause is confrontedwith the fact that the mechanisms that force disjoint reference are formally quite disparate.We arguethat the D/ f distinctionmakes available a moreprincipled treatment and helpsto derive the formal properties of reference-tracking systems (a notoriousproblem for standardbinding theory) without having to invoke construction-specific statements. Recallthat the categorical distinction between D and f correlateswith differences in their semanticsand their binding-theoretic status, as summarizedin (67). (67) Propertiesof D and f D f Semantics definite — Binding-theoreticstatus R-expression variable

Inprinciple, the D/ f distinctionholds of phrasal proforms (pro-DP vs.pro- fP), ofclitics (pro-D cliticsvs. pro- f clitics),and of X 0 -agreement(D 0 -agreementvs. f0 -agreement).In extend- ingour proposal to obviation and switch-reference, it will be necessary to appeal to the D/ f- agreementdistinction. More generally, in our analysis the binding properties of obviation and switch-referencereflect the D/ f distinction.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 433

(68) Formalproperties of reference-trackingsystems Obviation Switch-reference D‘‘obviative’’ ‘‘different-subject’’ f ‘‘proximate’’ ‘‘same-subject’’

We proposethat the disjoint reference effects found in obviationand switch-reference reflect the presenceof D-agreement,which predictably obeys Condition C andso triggers disjointness.

5.1Obviation Obviationis adiscourse-sensitivemechanism that distinguishes 3rd person participants from each other.For example,if averbhas more than one 3rd person argument, one of them is assigned proximatestatus, and all others are marked for obviation.In Plains Cree (Central Algonquian), obviationmarking occurs on overtDP argumentsin the form ofthesuffix -( w)a, (69).Obviation markingis alsoobligatory on possessed nouns, (70). (69)a. . ..eˆkotaka ˆ-waˆpamaˆ-cikiya ˆhciyiniwakoˆ hine ˆhiyaw- a. there COMP-see-3PL.PROX Blackfoot.PROX these Cree-OBV ‘...therethe Blackfoot caught sight of theCree.’ (adaptedfrom Long1999:96, (12)) b. Miyeˆ-w misatim-wa oˆhoˆkiseˆyini- wa eˆwakoˆ. give-3SG.PROX horse-OBV thatold.man- OBV this ‘He(PROX)gavea horse( OBV)tothe old man( OBV).’ (adaptedfrom Long1999:93, (3)) (70) aw ˆõ skweˆw o-naˆpem-a this woman.PROX 3.POSS-husband-OBV ‘thiswoman’ s husband’ (adaptedfrom Long1999:93, (4)) We proposethat obviation markers are categorically D; assuch, they respect Condition C. Thisimmediately predicts obligatory disjointness.

(71) DPi . . . DPj-OBV Thisproposal is consistent with Grafstein’ s (1984)analysis of obviation as a kindof ‘ ‘disjoint reference’’ marking.However, it departs from treatmentsthat view obviation marking as a Condi- tionB effect(Aissen 1997, De ´chaine1999). A ConditionB analysisincorrectly predicts that pronominalagreement should obligatorily encode obviation distinctions, and it must stipulate that fullDPs aremarked for obviation.Note that in (69) the verb bears pronominal agreement only for theproximate argument; it is the overt DP thatis marked for obviation.This asymmetry is capturedby a ConditionC analysis,which predicts that D-expressions— here full DPs— will be theprimary locus of obviation marking. A ConditionC accountalso generalizes to examples

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 434 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

suchas (70): obviation marking is obligatory on possessed nouns because they always involve thepresence of twoDPs. AConditionB accountdoes not predict this. AConditionC analysisof obviationalso leads us toexpect that when there is pronominal obviationagreement, it will be an augment of ‘ ‘normal’’ pronominalagreement (Blain 1997). Thisis becausepronominal agreement is f-agreement,and obviation agreement is D-agreement. (72)a. [V-stem]- f b.[V-stem]- f-D Theexpectation is fulfilled by examples such as (73a– b), where -w is f-agreement,and -yiwa isobviativeD-agreement, which is anextended form of f-agreement. (73) a. Seˆkiheˆ-w. frighten-3SG.PROX ‘He(PROX)frightenshim.’ (De ´chaine1999:64, (94a)) b. Seˆkiheˆ-yiwa. frighten-OBV.3 ‘He(OBV)frightenshim.’ (De ´chaine1999:64, (95)) D/f-agreementdiffer from eachother with respect to the discourse contexts that can felicitously hostthem. Proximate f-agreementis associated with the salient discourse referent and marks topiccontinuity (Goddard 1990, Russell 1991), (74a). Obviative D-agreement signals that the discoursereferent is distinctfrom theproximate topic, (74b).

(74) a. Kõˆweˆ-[w]i mistanask;koste ˆ-[w]i waˆkayoˆs-[a]j . go.home-3SG.PROX badger.PROX fear-3SG.PROX bear-OBV

‘Badgeri wenthome; he i was afraidof Grizzly j .’ (adaptedfrom Long1999:94, (7))

b. . . . , kaˆ-waˆpamaˆ-[t]i eˆh-kitoˆweˆhkwaˆmi-[yit]j . COMP-see-3SG.PROXCOMP -snore-3SG.OBV

‘. . . , hei(PROX) saw that hej(OBV)was soundasleep and snoring.’ (Long1999:98, (17), from P:80-23) Theclaim that obviation marking is a kindof D-agreementaccounts for bothgrammatically conditionedobviation (i.e., local disjoint reference) and discourse-conditioned obviation. This is aclassicalCondition C effect:DPs donot corefer.

5.2Switch-Reference The D/f distinctionalso provides insight into switch-reference systems, which are characterized byovert marking of disjointreference.

[Switch-reference]consists simply in the fact that a switchin subject or agent . ..isobligatorily indicatedin certain situations by a morpheme,usually suffixed . ..(Jacobsen1967:249)

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 435

For example,Mojave (Yuman) encodeswhether the subject of a dependentclause is coreferential ornoncoreferentialwith the subject of amainclause, as in(75). Obligatory coreference is marked bysame-subjectagreement; obligatory noncoreference is marked by different-subjectagreement. (75)a. nya-isvar- k iima-k when-sing-SS dance-TNS

‘When hei sang, hei danced.’ b.nya-isvar- m iima-k when-sing-DS dance-TNS

‘When hei sang, hej danced.’ (Munro1980:145, (4); as cited in Stirling 1993:3, (3)) Theessence of ourproposal is that different-subject agreement is D-agreement,while same- subjectagreement is f-agreement.This categorical distinction is supported by themorphosyntactic andbinding properties of different-subjectversus same-subject markers. 5.2.1Morphosyntactic Evidence Inseveral switch-reference languages there is a transparent morphologicalrelation between different-subject and same-subject markers. This is illustrated in (76)for Amele(Papuan). (76) Ameleswitch-reference markers (adaptedfrom Stirling1993:202) Singular Dual Plural Same DifferentSame DifferentSame Different subjectsubject subjectsubject subjectsubject

1 -ig -igin -u/À -wan -b -qon 2 -g -gan -si -sin -ig -gin 3 -i -n -si -sin -ig -gin

Suppletionsaside, we observea regularrelationship between the same-subject and different- subjectmarkers: the latter are augmented versions of the former. For example,in (76)the Amele 1stperson same-subject marker is -ig, andthe corresponding different-subject marker is -igin. In ourtheory pro-DP containspro- fPasa subconstituent,so we predictthat D-agreement (different- subjectmarking) should likewise contain f-agreement(same-subject marking) as a subpart.This predictionis borneout by the data. We nowshow how the binding properties of these two types ofagreementfollow from theircategorical status as D and f. 5.2.2Switch-Reference Is Mediatedby Tense As D-agreement,different-subject markers are predictedto be obligatorilydisjoint in reference.This is preciselythe nature of different-subject marking.Relative to another R-expression, D-agreement predictably respects Condition C and thereforeis disjoint in reference from anyother antecedent, as shown in (77). 2 6

26 We abstract away fromthe linear order of thedependent clause, whichusually precedes themain clause inswitch-

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 436 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(77) CP

C TP

[DP]i T9

T VP

VP TP

[D-agr]j However,by itself the D-agreement analysis does not predict the subject orientation of switch-referencemarkers. Unlike obviation marking, which can mark disjoint reference of an argumentindependent of grammaticalfunction, one of thecharacteristics of switch-reference is thatit isrestricted to subjects;that is, there is no distinct-object or same-objectmarking in switch- referencelanguages. As pointedout by Stirling (1993), switch-reference depends on tense, and we arguethat this tense dependency is responsiblefor thesubject orientation. Tosee how tense dependency derives subject orientation, consider how f-agreementwill workin such a system.If same-subjectmarkers are f-agreement,we expectthat they can be bound,and this is so. Two properties remain to be accounted for. First, why is f-agreement obligatorilycoreferential with an argumentin the main clause? Second, why must that argument beasubject?We showthat both of theseproperties reduce to obligatoryoperator binding of f- agreementand that the relevant A ¯ -operatoris Tense(Finer 1985, Stirling 1993). Thus, the licensing contextfor same-subject f-agreementis shown in (78).Reading the tree in (78)from thebottom up,the lower Tense operator A ¯ -binds the f-agreement.The same-subject effect is the result of aconspiracy.First, the lower Tense operator is anaphoricallydependent on the matrix Tense. 2 7 Second,each subject is coindexedwith its respective Tense via specifier-head agreement, thereby derivingthe subject orientation. Third, by transitivity the matrix and nonmatrix subjects are coindexed,resulting in obligatory coreference.

reference systems. We assume thatthe dependent TP raises tosentence-initial position, and we leave openthe nature of themechanism thattriggers this displacement. What is crucial forour analysis is that the matrix Tense c-commands the dependentTense. Evidence for a c-command relationstems fromthe fact thatthe tense ofthedependent clause is anaphoric onthetense ofthemain clause. See Stirling1993 for further discussion. 27 Ouranalysis predicts that a necessary andsufficient condition for same-subject markingis that the tenses ofthe matrixand nonmatrix clauses becoreferential. Thisis consistentwith Stirling’ s (1993)observation that same-subject markingdoes not always encodestrict coreference.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 437

(78) CP

C TP

[DP]i T9

Top VP

VP TP

Top [f -agr]i Thisanalysis generalizes to different-subject D-agreement, which is predictably subject to ConditionC, hencethe disjoint reference effect. In addition, we predictthat the disjoint reference forcedby D-agreementshould go handin hand with the independence of thelower Tense operator relativeto the matrix tense. This captures an important difference in the temporal properties of different-subjectand same-subject marking. 2 8 Remarkingon this property, Stirling makes the followingobservation:

Whereabsolute tense distinctions are marked, they are more likely to be marked on DS [different- subject]morphemes than on SS [same-subject]morphemes, which reflects a generaltendency for DS markedclauses to showa greaterresemblance to independent clauses than SS markedones. (Stirling 1993:43)

Byhypothesis, different-subject markers areD-agreement. As such,they have the status of R- expressions,which are referentially independent. In our analysis of switch-reference,and consis- tentwith Stirling’ s generalization,referential independence is parasitic on temporalindependence. If true,this leads us toexpectthat in switch-referencelanguages that have absolute tense distinc- tions,DPs willbe temporally independent (see Musan1995, Demirdache 1997b for relateddiscus- sion).This remains to be confirmed. Our analysisalso sheds light on oneof thedifferences between switch-reference and obvia- tion.While switch-reference is subjectoriented, obviation is not.For us,the subject orientation ofswitch-referenceis a by-productof thefact that it isparasitic on tense.If correct,this predicts

28 Anotheranalysis of switch-reference thatinvokes binding theory is proposedby Finer (1985), who extends the anaphor/pronoundistinction to the A ¯ -domainand treats same-subject markers as A ¯ -anaphorsand different-subject markers as A¯ -pronouns.The advantage of ouranalysisis that it accountsfor the fact thatswitch-reference istemporally conditioned, andalso derives the morphologically transparent relation between different-subjectmarkers (as D-agreement) andsame- subjectmarkers (as f-agreement). InFiner’ s analysisboth of these propertiesare accidental.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 438 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

thatin asystemthat is not tense dependent, disjoint-reference marking should cease to be subject oriented.We suggestthat this is preciselywhat happens with obviation agreement: it is simply disjointreference of twoDP argumentsindependent of their grammatical function.

5.3Reference-Tracking Systems: Summary Recognizingthe D/ f distinctionyields promising results for theanalysis of reference-tracking systems.Obviative systems are the effect of the categorical difference between D and f. Switch- referencesystems also exploit the D/ f distinction,but are subject to an additional restriction: theymust be operatorbound by Tense,which in turnderives the subject orientation of different- subjectand same-subject marking. (79) Formalproperties of reference-tracking systems Obviation Switch-reference D ‘‘obviative’’ ‘‘different-subject’’ w ‘‘proximate’’ ‘‘same-subject’’ Restriction — operatorbound by Tense Consequencen/ a subjectorientation

6Conclusionand Prospects Thecentral claim of the proposed analysis is that the notion ‘ ‘pronoun’’ isnot a primitive.We haveargued —onthebasis of evidencefrom predicate/argumentasymmetries, distributional tests, andbinding-theoretic properties— that it isnecessaryto differentiate at least three types of pro- forms:pro-DP, pro- fP,andpro-NP. Thesethree proforms stand in atransparentmorphological relationto eachother in that pro-DPs include fPs and/orNPs assubconstituents.This provides asolutionto theproblem that arises when pronouns are uniformly treated as DPs. Aconsequence ofadopting a finer-grainedsyntax for proformsis that it becomes possible to formalize the similaritiesand differences between obviation and switch-reference. Previousanalyses have argued for theexistenceof differentpronoun types, with most positing aDP shell,making it impossible to distinguish pronouns on the basis of their external syntax. Toourknowledge, the only other proposal that distinguishes pronouns from eachother in terms oftheircategorical identity is that of Cardinaletti and Starke (1999), who identify three pronoun types:strong pronouns, weak pronouns, and clitic pronouns. Tentatively, we suggestthat Cardina- lettiand Starke’ s proposalcan be reformulated in ourapproach in the following way. Their three pronounclasses might all be pro- fs,differing from eachother only internally. Accordingly, ‘‘strongpronouns’ ’ mightbe fPwiththe NP constituentproviding the range, (80a); ‘ ‘weak pronouns’’ mightbe phrasal fPs withno internalstructure, (80b); and ‘ ‘cliticpronouns’ ’ might be simplex fs, (80c).2 9

29 Logophoricpronominal systems (e.g.,Yoru ‘ba´)may instantiatesuch tripartite f-systems; see De´chaine and Wiltschko,in press.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 439

(80) a. f P b. f P c. f

f NP f

N Sincewe identify(at least) three syntactic layers within nominal expressions, each one of whichcan occur independently, we expectthat this should generalize beyond pronominals. In particular,the samethree-way distinction that holds of proforms should also hold of othernominal types,including reflexives, pro, agreement, and full XPs. Regardingthe possibility that the D/ f/Ndistinctionholds of reflexives, note that at aconcep- tuallevel our treatment of proforms is similar in spirit to analyses of anaphora that distinguish se-anaphorsfrom self-anaphorson the basis of their morphosyntax (e.g., Burzio 1991, Reinhart andReuland 1993, Safir 1996).Both approaches take as a startingpoint the idea that elements suchas ‘ ‘anaphors’’ and‘ ‘pronouns’’ donot constitute a uniformsyntactic class and that their binding-theoreticproperties follow from theirmorphosyntax. On independent grounds we have arguedthat Romance se-anaphorsare f. As for self-anaphors,our proposal is consistentwith two possibleanalyses: either they are DPs thatare ‘ ‘anaphorized’’ (Pica1987), or theyare nominal constantscontained within a DPshell. As for pro,in principle we expect(at least) three types: pro’ s thathave the syntax and semanticsof DP, of fP,orof NP. Tomioka(2000) observes that Japanese null pro has a wide rangeof semantic functions, including referential pro, bound variable pro, indefinite pro, and property-denotingpro. The existence of these different types of pro is expectedif proinstantiates differentsyntactic categories. In such a theoryit is not necessary to invoke type-shifting rules. Rather,the semantics can be read directly from thesyntax. Regardinginflectional agreement, again we expect(at least) three kinds: agreement that has thesyntax and semantics of D,of f,orofN. We havealready shown evidence for distinguishing D-agreementfrom f-agreement(section 5). N-agreement is arguably attested in Plains Cree in theform ofindefinitenessagreement, which is incomplementary distribution with incorporated nominalconstants (Hirose 2000). Finally,full XPs shouldalso show the D/ f/Ndistinction.We arguedin section4 thatwhereas Englishdefinite articles are associated with a DPstructure,French articles are introduced at the fPlevel,with predictable syntactic and semantic consequences. Within English there are reasons tothink that indefinites ( a girl)andbare plurals ( girls)instantiate fP.For example,they can functionas botharguments and predicates. 3 0

30 Stowell(1989) similarly observes that a nominalpredicate may bea nounby itself, a projectionof N(our fP), or a full DP.

(i)They elected her[ N president].

(ii)They consider him [ f P a rascal].

(iii)They consider her [ DP thestrongest candidate ].

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 440 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

(81) a. [fP Astronauts] A R G take risks.

b. [fP Anastronaut] A R G floatedacross the TV screen.

(82)a. John and Mary are [ fP astronauts] P R E D .

b. Mary is [f P anastronaut] P R E D . Thelayered analysis that we haveproposed for pronouns—and that possibly extends to all nominalexpressions— has its counterpart in the verbal domain. It is wellestablished that CP, IP, andVP allfunction as independent categories. In light of this,it would be surprisingif evidence were notfound for layeredsyntax in thenominal domain as well. Once the D/ f/Ndistinctionis recognized,then a numberof seeminglyunrelated properties of pronounsfall into place.

References Abney,Steven. 1987. The English noun phrase in itssentential aspect. Doctoral dissertation, MIT, Cambridge, Mass. Aissen,Judith. 1997. On the syntax of obviation. Language 73:705–750. Aoun,Joseph, and Norbert Hornstein. 1992. Bound and referential pronouns. In Logicalstructure and linguisticstructure, ed.by C.-T. James Huang, 1– 23. Dordrecht: Kluwer. Baggaley,Valerie. 1998. The syntactic category of pronouns. Master’ s thesis,University of Calgary. Bernabe´,Jean.1983. Fondal-natal:Grammaire basilectale approche ´e des cre´olesguadeloupe ´en et marti- niquais. 3vols.Paris: L’ Harmattan. Blain,Eleanor. 1997. Wh-constructionsin Ne ˆhiyaweˆwin(Plains Cree). Doctoral dissertation, University of BritishColumbia, Vancouver. Borer,Hagit. 1994. On the projection of arguments. In Functionalprojections, ed.by ElenaBenedicto and JeffreyRunner, 19 –48.University of MassachusettsOccasional Papers in Linguistics17. Amherst: Universityof Massachusetts, GLSA. Bouchard,Denis. 1984. Onthe content of emptycategories. Dordrecht:Foris. Burzio,Luigi. 1991. The morphological basis of anaphora. Journalof Linguistics 27:81–105. Cardinaletti,Anna. 1994. On the internal structure of pronominalDPs. TheLinguistic Review 11:191–219. Cardinaletti,Anna, and Michal Starke. 1999. The typology of structural deficiency: A casestudy of three classesof pronouns. In Cliticsin the languages of Europe, ed.by Henk van Riemsdijk, 145– 233. Berlin:Mouton. Chung,Sandra. 1989. On the notion ‘ ‘nullanaphor’ ’ inChamorro. In Thenull subject parameter, ed. by OsvaldoJaeggli and Kenneth Safir, 143– 184. Dordrecht: Kluwer. Corne,C. 1977. SeychellesCreole grammar. Tu¨bingen:Narr. Corne,C. 1988.Mauritian Creole reflexives. Journalof Pidginand Creole Languages 3:69–94. Davis,Henry, I.-J. Sandra Lai, and Lisa Matthewson. 1997. Cedar roots and singing detectives: Attributive constructionsin Englishand Salish. In Proceedingsof the32nd International Conference on Salish andNeighbouring Languages, 291–316. Port Angeles, Wash.: Peninsula College. De´chaine,Rose-Marie. 1993. Predicates across categories: Towards a category-neutralsyntax. Doctoral dissertation,University of Massachusetts,Amherst. De´chaine,Rose-Marie. 1999. What Algonquian morphology is really like: Hockett revisited. In Papers from theWorkshop on Structure and Constituency in Native American Languages, ed.by Leora Bar-el, Rose-MarieDe ´chaine,and Charlotte Reinholtz, 25 –72.MIT OccasionalPapers in Linguistics 17. Cambridge,Mass.: MIT, Department of Linguistics and Philosophy, MITWPL. De´chaine,Rose-Marie, and Victor Manfredi. 1994. Binding domains in Haitian. NaturalLanguage & Linguis- tic Theory 12:203–257.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 DECOMPOSINGPRONOUNS 441

De´chaine,Rose-Marie, and Martina Wiltschko. 2001. On pro-nouns and other ‘ ‘pronouns.’’ In Proceedings oftheNP-DP Conference, ed.by Martine Coene and Yves D’ Hulst.Amsterdam: John Benjamins. De´chaine,Rose-Marie, and Martina Wiltschko. In press. A modularapproach to reference-tracking. Ms., Universityof British Columbia. Demirdache,Hamida. 1997a. Condition C. In Atomismand binding, ed.by Hans Bennis, Pierre Pica, and JohanRooryck, 51 –87.Dordrecht: Foris. Demirdache,Hamida. 1997b. On the temporal location of predication types: The role of determiners in LillooetSalish. In Proceedingsof the 16th West Coast Conference on Formal Linguistics, ed. by EmilyCurtis, James Lyle, and Gabriel Webster, 129 –144.Stanford, Calif.: CSLI Publications. Diesing,Molly. 1992. Indefinites. Cambridge,Mass.: MIT Press. Di Sicullo,Anna Maria, and Edwin Williams. 1987. Onthe definition of word. Cambridge,Mass.: MIT Press. Finer,Daniel. 1985. The syntax of switch-reference. LinguisticInquiry 16:35–55. Galloway,Brent. 1993. Agrammarof UpriverHalkomelem. Berkeley:University of CaliforniaPress. Goddard,Ives. 1990. Aspects of thetopic structure in Fox narratives: Proximate shifts and the use of overt andinflectional NPs. InternationalJournal of AmericanLinguistics 56:317–340. Grafstein,Ann. 1984. Argument structure and the syntaxof anon-configurationallanguage. Doctoral disserta- tion,McGill University, Montre ´al. Hirose,Tomio. 2000. Origins of predicates: Evidence from Plains Cree. Doctoral dissertation, University ofBritishColumbia, Vancouver. Jacobsen,W. 1967.Switch-reference in Hokan-Coahuiltecan. In Studiesin southwestern ethnolinguistics, ed.by Dell Hymes and William Bittle, 238– 263. The Hague: Mouton. Katada,Fusa. 1991. The LF representationof anaphors. LinguisticInquiry 22:287–313. Kayne,Richard S. 1975. Frenchsyntax: The transformational cycle. Cambridge,Mass.: MIT Press. Kayne,Richard S. 2000. Parametersand universals. Oxford:Oxford University Press. Koster,Jan. 1997. Anaphora and the uniformity of grammar.In Atomismand binding, ed.by HansBennis, PierrePica, and Johan Rooryck, 235– 250. Dordrecht: Foris. Kuipers,Aert. 1974. TheShuswap language: Grammar, text, dictionary. TheHague: Mouton. Kuroda,S.-Y. 1965. Generative grammatical studies in the Japanese language. Doctoral dissertation, MIT, Cambridge,Mass. Lai,I.-J. Sandra. 1998. The grammar and acquisition of Secwepemctsõ ´nindependentpronouns. Master’ s thesis,University of British Columbia, Vancouver. Long,Michelle. 1999. Obviation and clausal relations in Plains Cree. In Papersfrom the Workshop on Structureand Constituency in Native American Languages, ed.by Leora Bar-el, Rose-Marie De ´- chaine,and Charlotte Reinholtz, 91 –100.MIT OccasionalPapers in Linguistics 17. Cambridge, Mass.:MIT, Department of Linguisticsand Philosophy, MITWPL. Longobardi,Giuseppe. 1994. Reference and proper names: A theoryof N-movementin syntax and Logical Form. LinguisticInquiry 25:609–665. Matthewson,Lisa. 1998. Determinersystems and quantificational strategies: Evidence from Salish. The Hague:Holland Academic Graphics. Munro,Pamela. 1980. On the syntactic status of switch-reference clauses: The special case of Mojave comitatives.In Studiesof switch-reference, ed.by PamelaMunro, 144– 159. UCLA Papersin Syntax 8.LosAngeles, Calif.: UCLA, Departmentof Linguistics. Musan,Renate. 1995. On the temporal interpretation of noun phrases. Doctoral dissertation, MIT, Cambridge, Mass. Noguchi,Tohru. 1997. Two typesof pronouns and variable binding, Language 73:770–797. Pica,Pierre. 1987. On the nature of the reflexivization cycle. In Proceedingsof NELS 17, ed. by Joyce McDonoughand Bernadette Plunkett, 483 –499.Amherst: University of Massachusetts,GLSA.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021 442 ROSE-MARIEDE´ CHAINEANDMARTINAWILTSCHKO

Pollock,Jean-Yves. 1998. On the syntax of subnominal clitics: Cliticization and ellipsis. Syntax 1:300–330. Postal,Paul. 1966. On so-called pronouns in English. In Modernstudies in English, ed.by David Reibel andSanford Schane, 201– 223. Englewood Cliffs, N.J.: Prentice-Hall. Postal,Paul. 1994. The ungrammaticality of subnominal en inFrench object-raising constructions. Linguistic Inquiry 25:179–186. Reinhart,Tanya. 1986. Center and periphery in the grammar of anaphora. In Studiesin the acquisition of anaphora,vol. 1, ed.by BarbaraLust, 123 –150.Dordrecht: Reidel. Reinhart,Tanya, and Eric Reuland. 1993. Reflexivity. LinguisticInquiry 24:657–720. Ritter,Elizabeth. 1995. On the syntactic category of pronouns and agreement. NaturalLanguage & Linguistic Theory 13:405–443. Russell,Kevin. 1991. Obviation as discoursestructure in a SwampyCree aˆcimowin. In Papersof theTwenty- SecondAlgonquian Conference, ed.by WilliamCowan, 320 –335.Ottawa: Carleton University. Safir,Kenneth. 1996. Semantic atoms of anaphora. NaturalLanguage & LinguisticTheory 14:545–589. Seiter,William J. 1980. Studiesin Niuean syntax. New York:Garland. Selkirk,Elisabeth. 1984. Phonologyand syntax: The relation between sound and structure. Cambridge, Mass.:MIT Press. Sportiche,Dominique. 1986. Zibun.Linguistic Inquiry 17:369–374. Stirling,Lesley. 1993. Switch-referenceand discourse representation. Cambridge:Cambridge University Press. Stowell,Tim. 1989.Subjects, specifiers and X-bar theory. In Alternativeconceptions of phrase structure, ed.by Mark Baltin and Anthony Kroch, 232 –262.Chicago: University of ChicagoPress. Sugamoto,Nobuko. 1989. Pronominality: A noun-pronouncontinuum. In Linguisticcategorization, ed. by RobertaL. Corrigan,Fred Eckman, and Michael Noonan, 267 –329.Amsterdam: John Benjamins. Tomioka,Satoshi. 2000. What can semantics say about pro-drop?Ms., University of Delaware,Newark. Vergnaud,Jean-Roger, and Maria Luisa Zubizarreta. 1992. The definite determiner and the inalienable constructionsin French and in English. LinguisticInquiry 23:595–652. Williams,Edwin S. 1997.Blocking and anaphora. LinguisticInquiry 28:577–628. Wiltschko,Martina. 1998a. On the syntax and semantics of (relative) pronouns and determiners. Journal ofComparativeGermanic Linguistics 2:143–181. Wiltschko,Martina. 1998b. The syntax of pronouns and determiners: A cross-linguisticstudy. In Current researchon language and linguistics, ed.by Marion Caldecott, Suzanne Gessner, and Eun-Sook Kim,293– 320. UBC WorkingPapers in Linguistics. Vancouver: University of British Columbia, Departmentof Linguistics. Wiltschko,Martina. 2002. The syntax of pronouns: Evidence from Halkomelem. NaturalLanguage & LinguisticTheory 20:157–195. Wunderlich,Dieter. 1996. Constructing inflectional paradigms. Paper presented at the Workshop on Inflec- tion,Vienna, 15 February.

Departmentof Linguistics Universityof British Columbia E270,1866 Main Mall Vancouver,British Columbia V6T 1Z1 Canada [email protected] [email protected]

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/002438902760168554 by guest on 27 September 2021