[x- rivista di grammatica generativa anno 1998 n. 23 uni pressRIVISTA Di GRAMMATICA GENERATIVA Volume 23, anno 1998 Alessandra TOMASELLI Prefazione p.l Lidia LONZ! II controllo argomentale nelle costruzioni e- spletive passive p.3 Gloria COCCHI On true double object constructions in Tshi- luba p.19 Elisa DI DOMENICO Constraining checking: on «-features repre- sentation and their parametric variation p. 41 Manuel ESPANOL-ECHEVARRIA Feature mismatches in Greek Clitic Left- & Angela RALLI . dislocation Constructions p. 63 Franco BENUCCI Bilinguismo precoce @ acquisizione delle pe- rifrasi verbali e del criterio wh- in italiano p.81PREFAZIONE I volumi nr. 23 24 detla Rivista di Grammatica Generativa ospitano la rielabora- zione di dieci delle diciannove relazioni presentate in occasione del XXIV Incontro di Gramnmatica Generativa svoltosi presso la Facolti di Lingue e Letterature Straniere dell’Universita degli Studi di Verona nell’ultimo fine settimana del febbraio 1998. Come si evince dal programma dell” Incontro riportato in calce a questa breve prefa- zione, |’appuntamento annuale dei generativisti italiani ha richiamato, come di consue- to, Pattenzione di un buon numero di studiosi provenienti da universita straniere. IL lento ma costante processo di intemazionalizzazione dell’incontro ha riproposto una senee dh questioni gia emerse negli anni precedenti, quali, ad esempio: la scelta della/e Ingue “ufficiali” (la possibilita di presentare una relazione in italiano va mantenuta, tavorita o addirittura scoraggiata?); Ia possibilita di garantire uno spazio alle nuove leve, in particolare ai dottorandi, la cui presenza, perlomeno in relazione all’esperienza veronese, ha segnano una notevole diminuzione. Vopportunita di prevedere la presen- za di “invited speakers” di richiamo; i vantaggusvantaggi della definizione di un tema- guida per le proposte di relazione (sul modello GLOW, per intenderci). Inevitabilmen- te, dato il grado di informalita che tuttora caratterizza 1'IGG ed il suo carattere voluta- mente itinerante, queste scelte non possono che essere lasciate alla responsabslita della sede organizzatrice. Solo in questo modo, del resto, gli incontri continuetanno a rap- presentare fedeimente le diverse reaita in cui é possibile portare avanti la ricerca nell’ ambito teorico genefativista in seno all’accademia italiana. Una veloce scorsa al programma del "98 ci permette di cogliere il prevalere di tre tendenze fondamentali, nonostante l’eterogeneita dei temi affrontati: * Vadesione al programma minimalista (proposto da Chomsky in diverse occasioni a partire dalla fine degii anni ’80 e che ha caratterizzato, di fatto, gran parte della produzione generativista del!’ ultimo decennio); + Papprofondimento della prospettiva teorica aperta da Kayne con la sua “The Anti- symmetry of Syntax” (Cambridge/Mass., MIT Press, 1994); « il tentativo di verificare ed estendere la consolidata metodologia del “vecchio” modelio teorico a principi e parametri a campi relativamente nuovi per la sintassi generativa, quale, ad esempio, il mutamento linguistico in prospettiva diacronica, la dialettologia, l'acquisizione di lingue seconde con particolare attenzione al bi- linguismo infantile precoce.La forza teorica delle due nuove prospettive di ricerca aperte dal minimalismo dal principio di antisimmetria, seppur dichiaratamente compatibili per lo meno a livel- lo programmatico, ha di fatto determinato una spaccatura fra gli studiosi di grammati- ca generativa che si é in parte manifestata anche in occasione dell’ IGG di \ erona. Come é evidente anche solo limitando la nostra attenzione agli articoli ospitati in que- sti due numeri detia Rivista di Grammatica Generativa, la contrapposizione fra “mini- malisti” “antisimmetristi" non sembra perd destinata a concretizzarsi in un dibattito aperto su argomenti di indagine comuni (ad esempio la ridefinizione del concetto di parametro ¢ delle modalita del confronto interlinguistico); al contrario, l'adesione all’'una od all'altra prospettiva teorica sembra comporiare, di fatto, una distinzione a livello di dominio empirico di ricerca. I fenomeni di accordo/di concordanza di tratti (fra tutti, la flessione verbale, l’assegnazione di Caso, il movimento di tipo di X°), ad esempio, continuano ad essere trattati quasi esclusivamente in chiave minimalista. Le questioni relative ali'interpretazione dei sintagmi quantificati (in altre parole le que- stioni di Forma Logica) cosi come i fenomeni di movimento che ricadono sotto etichetta di “remnant movement”, fra cui il focus ed il WH-movement e in senso lato i fenomeni di scrambling, costituiscono, al contrario, il dominio empirico di ricerca pri- vilegiato da chi si muove nella prospettiva dell’ antisimmetria. Le conseguenze di questa “‘spaccatura” a tiveilo empirico non sono certo facil- mente prevedibili, né valutabili, per lo meno dalla sottoscritta, La fluidita delta si- tuazione attuale, del resto, non esclude affatto un ricompattamento a livello teorico. Non mi sembra casuale, infatti, l'assenza a tutt’oggi di manuali introduttivi all’uno od all’altro modello. Alessandra Tomaselli Universiti degli Studi di Verona Facolta di Lingue e Letterature Straniere Via Lungadige di Porta Vittoria, 41 37129 Verona 26-28 febbraio 1998 - AULA C ulIL CONTROLLO ARGOMENTALE NELLE COSTRUZIONI ESPLETIVE PASSIVE! Lidia LONZI 1 presente articolo indaga su una proprieta rilevabile in italiano nelle frasi passi- ve con pro espletivo: la generale incapaciti del DP associato al pro di fungere da antecedente nelle costruzioni a controtlo. Siccome in questo tipo di frasi l’associato presenta l'accordo col verbo, occorre spiegare perché I’attesa correlazione tra ac- cordo € controllo (Chomsky, 1995; Cardinalciti, 1997) non si realizza. Scopo del mio lavoro é suggerire le linge Iungo le quali tale spiegazione potrebbe essere ricer- cata senza intaccare i fondamenti della menzionata correlazione. 1. Il problema Secondo una recente osservazione di Chomsky (seminario, 1997), mentre nelle costruzioni espletive con verbo inaccusativo il controtlo argomentale é regolarmente accettabile, in quelle passive lo stesso tipo di controtlo non lo é, o tutt’al piit é mar: ginale. Questo, in effetti, é quanto si riscontra in italiano. Si considerino le frasi e- spletive (1a-c), dove il DP associato al pro espletivo dovrebbe controllare il PRO. Mentre (1b) e (1c), com verbo inaccusativo, sono accettabili, (1a), in forma passiva, non lo é: (1a) @ ininterpretabile. In (2), che presenta le versioni con soggetto prever- bale delle frasi in (1), il controlio @ invece regolare: (1) a, *pro sono stati presentati gli attori, PRO; sfilando sul palcoscenico b, pro si sono presentati gli attori; PRO; sfilando sul palcoscenico ¢, pro é caduto un attore; , Giovanni, PRO; salendo sul palcoscenico 2) . Gli attori; sono stati presentati PRO; sfilando sul palcoscenico . Gli attori si sono presentati PRO; sfilando sul palcoscenico Un attore, (Giovanni; é caduto PRO, salendo sul palcoscenico s 1 Ringrazio Paola Beninca ¢ Anna Cardinaletti, per i Joro dettagliati commenti e suggeri- ‘menti in fasi precedenti di questo lavoro.Lidia Lonzi Nel prendere in considerazione le frasi (a), occorre tenere presente che il sogget- to di una frase passiva pud controllare in un gerundio di predicato soltanto se si trat- ta di un gerundio con valore temporale (Lonzi, 1998: 90), ¢ che il valore temporale funziona bene se c'é un rapporte d'incidenza di un’azione puntuale in un'azione du- rativa rappresentata dal yerundio, come & pi evidente in (3)-(4), dove tuttavia si ri- trova il contrasto rilevato mn (1 )-(2): (3) a. *pro @ stato investito Giovanni/un attore; PRO; uscendo dal teatro b. pro é caduto Giovanni/un attore; PRO; uscendo dal teatro (4) a. Giovanni/un attore; é stato investito PRO; uscendo dal teatro b. Giovanni/un attore; ¢ caduto PRO; uscendo dal teatro In (a-b) si conferma che 'espletiva passiva non permette il controllo dell’associato a differenza di quella con verbo inaccusativo (da ora in poi espletiva “inaccusativa”). Data l'analisi delle frasi espletive (e, ovviamente, la teoria del con- trollo) proposta in Chomsky (1995, cap. 4), questo fenomeno chiede una spiegazio- ne. Secondo Chomsky, nelle frasi espletive si ha una salita a LF dell’associato, che verifica i suoi tratti in Agrs “La salita covert in Agrs pone i tratti dell’associato in una posizione strutturale che ha le proprieta formali essenziali di (Spec,Agrs) C1 aspettiamo percid che l’associato abbia le proprieti di (...] controllo che ha il sugget to.” (Chomsky, 1995: 273). 11 problema, dunque, € il seguente. Se a LF (V’insieme di tratti “p” ¢ di Caso del)l’associato sale in Infl per verificare il Caso e I’accordo, ai fini del controtlo le frasi in (1a) ¢ (3a) dovrebbero comportarsi come quelle in (2a) ¢ (4a), dove il DP pertinente € salito in posizione di soggetto prima di Spell-Out ¢ c-comanda il PRO. Ma é lo stesso Chomsky a farci osservare che questo non avviene. Il problema é Pinaccettabilita di (1a) e (3a) 2. Esempi marginali ed esempi inaccettabili Come (1a), anche (5a), con un‘infinitiva di scopo, & inaccettabile, perché Vinterpretazione intesa richiederebbe il controllo argomentale, che, appunto, nelle espletive passive non é accessibile (5) a, *pro é stato punito Giovanni, per PRO; essere di esempio a Paolo b, Giovanni é stato punito per PRO; essere di esempio a Paolo Che Vinaccettabiliti di (Sa) sia imputabile alla costruzione espletiva, dimostrato dall’accettabilita di (Sb), con soggetto preverbale. Anche se alcuni avverbiali, in particolare le infinitive precedute da senza, sembrano permettere i! controllo argomentale, questo avviene con risultati di stile trascurato o substandard. V. (6a) versus (6b):H controlto argomentale nelle costruzioni espletive passive (6) a. ?2pro sono stati trovati tre bambini senza PRO essere nemmeno stati cercati b, tre bambini sono stati trovati senza PRO essere nemmeno stati cercati Nelle frasi a controllo argomentale in (6), la variante con sozsetto lessicale preverbale @ decisamente migliore. Poiché gli avverbiali che producono questo tipo di risultati sono in genere le infinitive precedute da preposizione, da ora in poi prenderd in considerazione gli avverbiali al gerundio che rispondono a condi- zioni pitt definite, 3. Proposta di spiegazione Nella mia analisi, nell’espletiva passiva l’associato non sale in Infl perché non ha Caso nominativo, e pertanto non c-comanda il PRO nemmeno a LF. Si pud ren- dere ragione di questo fatto suggerendo che nelle costruzioni espletive il suffisso participiale non assorba il Caso accusativo. A LF sarebbe lo stesso participio ad segnare questo Caso all’oggetto (Pollock, 1981) che lo verificherebbe in Agro. A Spell-Out, invece, conforme alla Nominative Agreement Hypothesis di Cardi- naletti (1997: 522), l'accordo sarebbe stabilito con Passociato, in quanto il pro é in- trinsecamente sottospecificato per il Caso. L’espletivo forma una catena argomenta- Ie con l’associato ¢ in tal modo soddisfa I"EPP, come avviene in qualunque costru- zione espletiva con verbo inaccusativo (Chomsky, 1995: 273-274). Liintuizione che sta alla base della presente proposta é che, a LF, proprio la pre- senza di questo pro “sogzetto” possa bloccare il meccanismo del passivo, vale a dire Vassorbimento di usativo € di ruolo tematico esterno da parte della cosid- Getta morfologia “passiva", cosicch¢ il participio passato riacquista valore transitivo € il pro pud ricevere il ruolo tematico esterno. La possibile assegnazione di Caso accusativo da parte del participio passato é quindi un‘ipotesi cruciale. E stata avan- zata da tempo per il francese, pili precisamente per il passivo impersonale (Pollock, 1981), © successivamente confermata per varie lingue dalla letteratura sull’argomento (citata in Baker, Johnson & Roberts, 1989: 236, dove questa forma & deta “transitive passive” in considerazione dell’oggetto accusativo). E da notare, incidentalmente, che la proposta di Pollock sembra fare la corretta distinzione tra ‘associato” ¢ soggetto invertito ai fini della possibilita del controllo: solo il soggetto invertito mostra di poter controllare (Guéron, comunicazione personale2), Un’altra 2 Lo stesso sembra valere anche per I’italiano, nonostante non ci sia lo stesso comporta- mento risperto all'accordo, Per es.: (i) E stato riaccompagnato IL RAGAZZO, nella sua/propria stanza Gi) Sono stati presentati GLI ATTORI, PRO sfilando sul palcoscenico E da notare che in questi casi anche l'avverbiale & presupposto.Lidia Lonzi ipotesi cruciale per quest analisi @ che Agro sia presente anche nel passivo, in linea con gli assunti di Chomsky (1995), v. anche Cingue (1997). Un test standard utilizzabile per provare che siamo in presenza di un oggetto € non di un soggetto é I’uso del riflessivo proprio}. V. gli ess. in (7): (7) a. Il ragazzo @ stato riaccompagnato nella sua/propria stanza b. E stato riaccompagnato il ragazzo nella sua/propria stanza c. Abbiamo riaccompagnato il ragazzo nella sua/*propria stanza d. E arrivato il ragazzo con la sua/propria auto In (7a), con soggetto lessicale preverbale, il riflessivo & ammesso, mentre in (7) non lo @: l"associato sembra comportarsi come l’oggetto in (7c), Anche se dobsiamo ricordare che la salita a LF non é del tutto rilevante per il binding (Chomsky, 1995: 275; Cardinaletti, 1997: 525), I'espletiva inaccusativa (74), che contrasta con quella passiva in (7b), sembra confermare la validita del test se, come é legittimo assumere, lo status pertinente dei due complementi ¢ comparabile, Sinteticamente, in presenza di un pro “soggetto”: i) il ruolo tematico esterno viene assegnato al pro per il Criterio Theta; ii) Massociato pud fare checking di Caso accusativo in \y.,. iii) pro pud fare checking di Caso nominativo perché non « ¢ un DP che possa com- petere con esso (Cardinaletti, 1997: 531); iv) Vintero processo si conforma alia Generalizzazione di Burzio (1986). In particolare, mi sembra che nei passivi impersonali, o “passivi transitivi”, per salvare la generalizzazione di Burzio non sia necessario pensare che il ruolo temati- co esterno sia assegnato al suffisso passivo, come é suggento in Baker, Johnson, & Roberts (1989). Lipotesi avanzata pitt sopra che sia assegnato all’espletivo sarebbe preferibile perché pit motivata. Mente in Baker, Johnson & Roberts l’assegnazione del ruolo tematico esterno al suffisso participiale era essenzialmente giustificata dal fatto di essere in tal modo prevista in entrambe le forine di passivo, a partire da quella intransitiva, nella mia analisi l’assegnazione del ruolo tematico esterno al pro entrerebbe in una dinamica di tipo transitivo. Pid avanti, porterd un argomento a fa- vore di quest’analisi, costituito da un caso - il gerundio coordinato - in cui il pro e- spletivo sembra comportarsi come un argomento, Quéello che € singolare - quello che “offende” - nella mia proposta, ¢ che a de- terminare l’accordo del verbo sia un DP con Caso accusativo4, oppure, pitt corret- 3° Nella loro analisi dell’espletiva passiva in danese, condotta in un divers quadro teorico, Heltoft & Jakobsen (1995) adottano questo test per rafforzare la loro proposta di conside- rare il soggetto “unpromoted” come un vero e proprio oggetto.Jl controllo argomentale nelle costruzioni espletive passive tamente, che a ricevere Caso accusativo sia un DP che determina I’accordo del ver- bo. Poiché determina I’accordo, questo DP dovrebbe piuttosto essere un soggetto, secondo la regolare associazione accordo-Caso nominativo che si osserva in italiano (v. per es. Giorgi & Pianesi, 1997) Ma non é forse irragionevole assumere che men- tre i tratti della testa Agrs che esizono di essere checked, lo sono attraverso pro gra- zie all’ avvenuta trasmissione de: (ratt: dell’associato in un qualche modo prestabilito (Burzio, 1986; v. Chomsky, 1995: 392), il checking del Caso accusativo in Agro avvenga come processo a sé, parassita della struttura transitiva. In questo quadro, Vaccordo del verbo con I'associato é un fossile senza potere di attrazione, che do- cumenta l’esistenza dei due livelli: Spell-Out e LF. Nell’analisi appena proposta, dunque, l'associato riceve regolarmente Caso ac- cusativo: il suffisso passive non pud assorbirlo come non assorbe il ruolo tematico estemno, che viene assegnato al pro. Il checking dell"espletivo” & regolare (Caso nominativo ¢ tratti di accordo dell’associato, come nell’ inaccusativo), fatto appunto da uno stesso elemento: pro (Cardinaletti, 1997: 531); il checking in Agro da parte dell’ associato é altrettanto regolare, perché la derivazione si sfascerebbe per manca- to checking dell’accusativo. La conseguente regolare non salita dell’associato ne esclude la capacita di controllo. 4. Un argomento a favore: il controllo agentivo nei gerundi “coordinati” A sostegno della mia spiegazione, penso sia utile passare a considerare Pespletiva passiva dal punto di vista del controllo non pit: argomentale ma agentivo. Si vedano dunque (8a) e (9a), da confrontare come esempi particolari di controll agentivo (la cui natura specifica sara precisata pitt avanti) con gli inaccettabili (1a) ¢ 3a) pid sopra, a controllo argomentale. V. anche (8b) ¢ (9b),5 con soggetto prever- bale, in cui si osserva il regolare controllo agentivo, attribuibile al suffisso del parti- cipio passivo. Questo tipo di controllo opera limitatamente al dominio del VP (Lon 4) Forse una forma assimiabile a quella dell’espietiva passiva é la forma impersonale con si passivante € con ogzetto in sede che determina Paccordo (Qui, si mangiano i dolci al cioccolato). In questa forma, che non & riconducibile al passivo “morfologico” studiato da Belletti (1982) perché non ne & una semplice inversione, il si, conforme all’analisi di Cin- que, avrebbe caso nominativo € muolo tematico estermo (per uta posizione diversa v. Do- brovie-Sorin, 1998). L‘assimilazione all'espletiva passiva potrebbe servire a sormontare T"apparente paradosso” costituito dal fatto che anche il DP oggetto, sempre secondo Vanalisi di Cingue, non pud non avere caso nominativo (“dato che solo i sintagmi nomi- nativi possono determinare T’accordo in italiano”. Cinque, 1988: 558). Ma tutta Ja que- stione merita uno studio a parte 5 Alcuni parlanti preferiscono inserire una pausa prima dell’avverbiale, optando per un’intonazione staccata che non ne cambia perd la natura di predicato (Lonzi, 1991),Lidia Lonzi 2i, 1998, cap. 3), ¢ in (8b)-(9b) & appunto determinato, neli’ordine, dai valori di ma- niera e di scopo dell’avverbiale (8) a. pro sono stati presentati gli attori(,) PRO elencandone le interpretazioni pitt importanti b. Gli attori sono stati presentati; (dal regista;) PRO; clencandone le interpreta- zioni pit importanti (9) a. pro & stato punito Giovanni(,) per PRO vendicare Paolo b. Giovanni é stato punito; (dagli amici;) per PRO; vendicare Paolo ‘Ora, sappiamo che j gerundi “coordinati” (e, pid in generale, i gerund: aggumti a una projezione funzionale alta della frase, come per esempio quelli causali), presu- mibilmente per ragioni configurazionali non ammettono il controllo agentivo, come mostrano (10a)-(13a) (10) a, *Giovanni era stato punito, (dagli amici,), PRO; volendo in tal modo vendi- care Paolo (11) a. *Gli attori sono stati presentati; (dal regista,), PRO, avendo deciso di aiutarli (12) a. *Lo studente non era stato interrogato; (dalla commissione)), PRO; agendo correttamente (13) a. *Gli alunni sono stati contati, (dall’insegnante)), PRO; mettendosi poi a disposizione dei genitori ma quello che qui interessa ¢ che lo ammettono quando la frase passiva & appunto una frase espletiva, v. (10b)-(136): (10) b. pro; era stato punito Giovanni, PRO, volendo in tal modo vendicare Paolo (11) b. pro; sono stati presentati gli attori, PRO, avendo deciso di aiutarli (12) b. pro;é stato interrogato un solo studente, PRO, agendo correttamente osizione dei genitori (13) b. projsono stati conta gh alunni, PRO; mettendosi poi adi Le coppie minime costituite dalle frasi (a) vs. (b), costituiscono un contrasto su- seettibile di valutazioni diverse, su cui tornerd piti avanti. Qui mi basta aver mostra- to che le frasi (b) sono decisamente pitt accettabili delle frasi (a). Si noti che in (10a)-(13a) anche il controlio di frase é escluso per ragioni di plausibiliti semantica Viceversa in (14a) e (15a) pit sotto, dove si potrebbe pensare, erroneamente, che sia in atto il controllo agentivo, é l’evento denotato dalla frase matrice che costituisce Vantecedente del PRO, come dimostrano (14b)-(15b), inaccettabili_ per l'incompatibiliti del predicato gerundivo con un evento come soggetto:I contrailo argomentale nelle costruzioni espletive passive (14) a. cx (Certi errori erano stati ammessi (dal presidente), PRO; suscitando la disapprovazione generale = il che ha suscitato la disapprovazione generale b. cpi*(Certi errori erano stati ammessi; (dal presidente,)), PRO; avendo temu- to la disapprovazione generale (15) a. cri (Gli attori sono stati presentati dal regista), PRO; placando la curiosita del pubblico = i che ha placato la curiosita del pubblico b. ce*(Gli attori sono stati presentati; dal regista), PROy avendo intuito la cu- Tiosita del pubblico L*inaccettabilita di (10a)-(13a), a controllo agentivo, presumibilmente dovuta a ragioni configurazionali, e di (14b)-(15b), a controllo di frase, chiaramente dovuta a ragioni di plausibilita semantica (a parte la predicibile impossibiliti di quello agen- tivo), cospirano a dimostrare che l'accettabilita di (14a)-(1Sa) dipende dal controlio di frase ¢ che ’indicizzazione corretta del PRO in (14a)-(15a) € quella indicata, Nel- Ja mia analisi, il controllo di tipo arb, che pud applicarsi nel gerundio coordinato, & legittimato dal controllo di frase (v. Lasnik, 1988; Lonzi, 1998, cap. 5). Per es., in (15a), si pud intendere che sia un agente imprecisato, responsabile della decisione di far presentare gli attori (dal regista), a placare la curiosit’ del pubblico, ma come conseguenza del fatto che I’ha placata l’evento della presentazione Che le cose stiano cosi & confermato da (16}-(17), rispettivamente con soggetto preverbale e pro espletivo. In (16b), un predicato che ¢ incompatibile con un evento come soggctto rende Ja frase inaccettabile, a conferma che (16a) ¢ a controllo di fra se. Lo stesso contrasto, pero, non si osserva in (17), dove anche la frase (b) & accet- tabile, a conferma che Ja struttura espletiva mette in gioco un altro tipo di controllo: (16) a. Il magistrato é stato ammonito (dai politici), PRO sorprendendo tutti strato é stato ammonito (dai politici), PRO non avendo sopportato le sue cntuche (37) a. pro @ stato ancora ammonito il magistrato, PRO sorprendendo tutti b. pro é stato ancora ammonito il magistrato, PRO non avendo sopportato le sue critiche Mentre in (16a) la frase matrice pud essere if soggetto dell’azione espressa dal gerundio e quindi (16a) é accettabile perché ammette il controllo di frase, col conse- guente effetto puramente interpretativo di un controllo agentivo di tipo arb, in (160), in presenza del predicato a soggetto [+umano], la via di uscita del controllo di frase non fa senso: un fatto non pud “sopportare” alcunché, € la frase é inaccettabile. In (176), invece, dove la frase passiva una costruzione espletiva, anche la frase (b), oltre alla frase (a), € del tutto accettabile. In particolare, siccome in (176) il controllo del suffisso participiale & escluso per V'inaccettabilita di (16b), dobbiamo pensareLidia Lonzi che sia il pro a controllare, La stessa conclusione s’impone per gli ess, (10b)-(136). Nelle espletive passive, dunque, perché ci sia controllo non occorre if rispetto della condizione vista in (14)-(16), cioé che la frase principale abbia i requisiti semantici per essere il soggetto dell’avverbiale, Antecedente del PRO pud essere un soggetto genetico impersonale, nella mia analisi il pro espletivo, che, per la dinamica partico- lare suggenta in 3., assume status argomentale. La disponibilitd al controllo da parte del ruolo tematico estemo nella frase passiva con pro espletivo ¢ non in quella con soggetio preverbale, dev'essere ricondotta a una specifica differenza di posizione dell’argomento esterno nelle due forme di passivo. La conclusione pertinente e che pro dev'essere considerato un vero e proprio controllore nelle frasi con zerundio aggiunto a una proiezione funzionale alta che chiamano in gioco un antevedente c- comandante pid alto del suffisso in Voice. 5. Controllo del suffisso passivo e conirollo del pro espletivo Se lo stesso pro dell’espletiva, conforme all’analisi qui presentata, ha capacita di controllare nel gerundio coordinato in quanto {+arg], occorre pensare che possa con- trollare anche dentro all’avverbiale di predicato, Pit precisamente, nelle espletive passive (8a)-(9a), che qui ripeto, il controllo agentivo non sarebbe determinato dal suffisso participiale bensi dal pro: (8) a. pro sono stati presentati gli attori(,) PRO elencandone le interpretazioni pit importanti b. Gli attori sono stati presentati; PRO; elencandone le interpretaziom pil im- portanti (9) a. pro é stato punito Giovanni(,) per PRO vendicare Paolo b. Giovanni é stato punito; per PRO; vendicare Paolo La logica dell'analisi dett’espletiva passiva come “transitiva”, infatti, impone che il controllo agentivo operante nelle frasi (a) sia attribuito al pro, mentre_nelle frasi (b), con soggetto lessicale (qui volutamente escluso come antecedente), dev’essere il suffisso passivo a controllare, anche se, come si @ visto, pud farlo solo negli avverbiali di predicato presumibilmente per ragioni di tipo configurazionale.6 Questo dato del controllo agentivo in una struttura in cui non & previsto (fra 10b-13b), & complementare al dato segnalato da Chomsky del mancato controllo dell’associato in una struttura in cui é previsto. In questo caso abbiamo un DP che dovrebbe verificare il Caso nominativo e che non sale in Infl, nel caso illustrato in 6 In Baker, Johnson & Roberts (1989), tale spiegazione non sarebbe disponibile perché il suffisso & generato in Inf. 10I controllo argomentale nelle costruzioni espletive passive (J0b)-(13b) abbiamo un ruolo tematico estemo che controlla come se fosse in una posizione pit alta del suffisso passivo, convenzionalmente collocato in Voice. Pit precisamenie, date le generali assunzioni sul gerundio causale, come se fosse in Spec, di Agrs/T. Anche in (10b)-(13b), poiche il suffisso passive non pud controlla- re, & ragionevole pensare che il ruolo tematico esiemno sia assegnato al pro, che guindi non formerebbe una catena argomentale con |’associato. La catena argomen- iale pro-associato, propria delle costruzioni espletive, ne! passivo pud dunque non attivarsi. Vedremo pits avanti che vi é almeno un contesto particolare - a parte i casi di soggetto invertito - in cui tale catena si attiva. In questo caso si riescono a ottene- re delle coppie minime che militano a favore di quest’analisi, purtroppo affidata a giudizi difficili e talvolta sfuggenti7 6 Prime conclusioni Secondo l’analisi qui presentata, nell’interpretazione delle espletive passive si at- tiva un meccanismo che non permette la promozione dell’ associato alla posizione di soggetto che in definitiva rompe la catena argomentale espletivo-associato. La conseguente non salita dell’associato impedisce la sua capacita di controllo, che & invece evidente nelle espletive inaccusative, Alcuni aspetti del fenomeno sono spie- gabili alla luce delf’analisi delle espletive sviluppata in Chomsky (1995) e Cardina- letti (1997). Per es., se accettiamo }’analisi che solo I’espletivo non ambiguamente nominativo permette l'accordo cot verbo, dobbiamo anche accettare che pro non lo permetia (Cardinaletti, 1997), e che in tal caso ’accordo debba essere con Passociato, quale DP pitt prominente, Inoitre, @ importante ricordare che mentre la realizzazione morfologica, owvia- mente, pud solo aver luogo nella sintassi overt, il checking dei tratti pertinenti puo aver luogo a LF. In particolare, quello del Caso astratto potrebbe avvenire sempre € solo a LF (Chomsky, 1995: 67). Pertanto, o un tratto € checked nella sintassi overt, nel qual caso si ha anche la configurazione richiesta per il checking, 0 non io é, nel qual caso é sufficiente che la configurazione richiesta si abbia al livello pertinente, cioé a LF. Nel caso specifico delle frasi espletive sappiamo che, (i) se Ja realizzazione morfologica di Agreement é con i tratti “o” dell’ associato - € quindi, proprio perché si é in presenza di una costruzione espletiva, non c’é 7 Alcuni parlanti non condividono i giudizi cruciali dati in (10a)-(13a), perché accettano il controllo arb indipendentemente dalla sua riducibilit8 al controlio dell'evento, quindi in- dipendentemente dal tipo di predicato che pud anche selezionare un soggetto [-+umano}. Forse, nonostante la disparita dei giudizi, & possibile concludere che questa accettazione rientra in uno stile in qualche modo informale. M1Lidia Lonzi la configurazione richiesta - i tratti del DP pertinente, ovviamente, non sono checked nella sintassi overt (es.: pro si sono presentati gli attori), (ii) se la realizzazione morfologica di Agreement non & con i tratti “p” dell’ associato, il checking relativo avviene per altra via, pit: precisamente vie~ ne compiuto dall'espletivo, presumibilmente nella sintassi overt (es.: Hl a été arreté trois linguistes, Pollock, 1981); y. in particolare Cardinaletti (1997) per esempi precisi di accordo con l’espletivo; (iii) infine, come si @ visto, il checking del Caso da parte dell’ espletivo potrebbe forse avvenire anche quando la realizzazione morfologica di Agreement é con i tratti “p” dell’associato, se non si attiva il processo di salita a LF del DP perti- nente, F ’ipotesi presa qui in considerazione per le frasi espletive passive. Qui, si noti, il checking avverrebbe obbligatoriamente a LY In conclusione, si ha liberti di realizzazione morfologica (dei tratti di Agree- ment) solo nel caso che i tratti “ip” non siano checked dal DP pertinente (quindi il soggetto sia postverbale, Guasti e Rizzi, 1998). In tal caso questi tratti o sono che- cked a LF (dal DP pertinente) - y. Je frasi espletive con verbo inaccusativo, dove la realizzazione morfologica avviene per trasmissione all’espletivo dei tratti dell’associato, ma it checking relativo avviene per raising a LF dell"associato (Chomsky, 1995:273), 0 sono checked dall"espletivo, indipendentemente dal fatto che li riceva 0 no dallassociato - v. le frasi espletive passive dell’italiano e, rispettivamente, quelle de} francese. La predizione relativa, in accordo con la letteratura sull’argomento ¢ in particolare Chomsixy (1995), Cardinaletti (1997), & che, se esistono costruzioni espletive di forma passiva in cui l’associato determina l’accordo del verbo e, crucialmente, ha Caso n0- minativo, deve valere la correlazione accordo-controllo. Quindi, se I'associato é c ramente nominativo, ci deve essere raising dei tratti. In caso contrario, si applica la strategia alternativa qui proposta, cio? l'associato riceve Caso accusativo. 7 L’associato nominative determina il controllo 7.1. Un’indiretta conferma di quest’analisi & costituita dai casi in cui associat & una forma pronominale nominativa, Come era desiderabile, le capaciti di controllo del pronome nominativo nell’espletiva passiva in (18b) sono le stesse illustrate in (18a) col pronome in posizione preverbale, e questo prova la correttezza della corre- Jazione tra accordo ¢ controllo, che nella sua motivazione passa proprio attraverso il Caso nominativo: (18) a. [o; sono stata visitata, PRO,.; tenendo le mani sopra alla testa b. pro; sono stata visitata 1O;, PRO; tenendo le mani sopra alla testa ©. %pro; sono stata visitata io; PRO; tenendo le mani sopra alla testaI controllo argomentale nelle costruzioni espletive passive D'altro lato, poiché la forma nominativa postverbale sembra possibile soltanto con verbo presupposto, & presumibile che questa sia in grado di controllare in quanto soggetto invertito (v. nota 1) € non come puro associato, II dato illustrato in (18b), in effetti, potrebbe essere una prova determinante a favore della mia analisi solo se il soggetto invertito yenisse analizzato nella sua posizione di ogget- to (v. in proposito Cardinaletti, 1995), 0 se il pronome nominativo potesse co- munque entrare in una costruzione informativamente non marcata, come & sugge- rito in (18c), un'ipotesi difficile da documentare perché, mentre una differenza intonativa & possibile tra (18b) e (18c) (nel senso indicato dal carattere maiusco- Jo), la presupposizione di frase non sembra cambiare. Qui, tuttavia, adotto lanalisi tradizionale secondo cui il soggetto invertito & in una posizione di ag- giunto e viene reinterpretato in posizione soggetto, cosi da spiegare In sua diversi- ta di comportamento rispetto all’associato nel controllo argomentale, Pertanto, pit che una prova a favore della mia analisi, il dato in (18) potrebbe essere un’indiretta conferma della sua correttezza. Le stesse considerazioni vanno fatte per il predicibile comportamento del pro espletivo in questi stessi casi. In (19), con gerundio coordinato, si assiste al blocco della sua capacita di controllo, che non si afferma piti come possibile in contrasto con (19a) a seggetto preverbale (in cui sappiamo che il controllo agentivo & impos- sibile, y. 10-13) (19) a, *Io sono stata visitata;, PRO, mettendosi poi a disposizione degli altri pa- zienti b. *pro sono stata visitata 1O/io(,) PRO. mettendosi poi a disposizione degli altri pazienti U1 pro, qui, forma una catena argomentale col pronome nominativo, che controlla come se fosse in posizione soggetto, in accordo col dato in (18) In conclusione, le espletive passive con pronome nominativo hanno il grande in- teresse di presentare esiti esattamente opposti a quelli finora osservati nelle espletive passive in generale. Pit precisamente, controllo argomentale dell’associato nel ge- rundio di predicato, v. (18b), mancato controllo agentivo (del pro) nel gerundio coordinato, v. (9b). Questo dato é coerente con l’analisi qui proposta per le espleti- ye passive in italiano, secondo cui l’associato non sale in Infl perché non ha Caso nominativo, 8 Si noti che il dato in (19) corrisponde a quanto si osserva nel gerundio di predicato per il controllo agentivo, che non costituisce pitt un’altemativa possibile al controilo argomen- tale se a determinare il controllo argomentale ¢ un pronome nominativo (Lonzi, 1998, cap.3), come si pud vedere, per es., in (18a). Ringrazio Anna Cardinaletti per avermi sug- gerito questo tipo di esempi 13Lidia Lonzi 2. Una conferma specifica, in un contesto informativamente non marcato, potrebbe venire dalla capacita di controllo dell’associato nelle forma di passivo in cui viene passivizzato il verbo aspettuale, detta qui “passivo a ristrutturazione” o anche, per semplicita, “passivo lingo”. Secondo Cingue (1998), questa sorta di passivo a ristrutturazione & possibile so- lo con quei verbi (come finire e iniziare) che possono fare checking del loro specifi- co tratto aspettuale restando al di sotto della proiezione di Voice, dove si suppone che approdino i quanto participi passivi per il checking morfologico. E opinione condivisa da molti linguisti che in italiano i! participio passato possa salire in una Proiezione piu alta, ma questo non avrebbe conseguenze nel rendere accessibile la passivizzazione ad altri verbi aspettuali. Il fatto che questa non sia ammessa con i verbi il cui tratto aspettuale sarebbe da verificare in una proiezione pid alta di Voice, suggerisce che questo tipo di tratti debba essere verificato prima di Voice. Nella struttura data in (20) (v. Cinque, 1997), un Aspetto completive(I1) @ con- trapposto a un Aspetto completive({) (o anche a un Aspetto terminative) pid in alto di Voice, cosi come un Aspetto inceptive({I) ¢ contrapposto a un Aspetto incepti- ve(1): gli uni - completive(I[) ¢ inceptive(IT) - riguardano un punto naturale di fine o @inizio di un processo, gli altri - completive(T) € inceptive(1) - un punto arbitrario. Nell’analisi di Cinque, é l'ordine delle teste aspettuali rispetto a Voice che spiega la possibilita di ristrutturazione che, nel passivo, certi verbi presentano ¢ non altri: (20) Voice >... Aspeontinuativett! ASPinceptivec) /Andative/ ASPcomptesivet) (V) L*impossibilita di frasi con il valore completive({), v. (21) con spingere vs. (22) con costruire (= 20b e 20a in Cinque, 1998), dimostra che la proiezione aspettuale pertinente é piti alta di Voice: (21) *Lamacchina fu fina di spingere alle 5 (22) Quella casa fu finita di costruire nel 1950 Ora, per quanto concerne la costruzione espletiva del passivo lungo, il control- lo argomentale sembra possibile. I dati in (23)-(24) dimostrano infatti che Vassociato controlla anche dentro I’avverbiale gerundivo (qui di tipo coordinato) Le frasi a ristrutturazione (23a-24a) danno risultati migliori rispetto alle semplici frasi passive (23b-24b), che confermano invece I’incapacita dell’associato a con- trollare: (23) a. %pro furono finite di costruire le nostre due case, PRO portando il loro con- tributo al paesaggio circostante b. *pro furono costruite le nostre due case/ alcune case, PRO portando il loro contributo al paesaggio circostanteJi controllo argomentale nelle costruzioni espletive passive (24) a. %pro fu iniziata a costruire una chiesa, PRO stampandosi con la sua mole nelPimmaginario di tutti b. *pro fu costruita una chiesa, PRO stampandosi nell'immaginario di tutti Siamo indotti a pensare che nei passivi a ristrutturazione il suffisso passivo possa assorbire regolarmente I’argomento esterno € il Caso accusativo. Ii control- Jo dell'associato nel passivo Jungo sarebbe finalmente possibile in quanto qui Poggetto, per ricevere Caso, dovrebbe salire in Infl per il checking del Caso no- minatiyo. Il passivo tungo, dunque, non solo richiederebbe che il checking aspet- tuale del participio sia prioritario rispetto a quello in Voice, ma anche che il parti- cipio passato obbedisca alle condizioni proprie del passivo, un dato che sembra militare a favore dell’effettiva rilevanza della proiezione pertinente in questa par- ticolare formazione. Pi esplicitamente, quest’ analisi implica, in manicra piuttosto ragionevole, che il passivo a ristrutturazione richieda la regolare applicazione del meccanismo del passivo. I contrast illustrati nelle frasi (23)-(24), dove le forme a ristrutturazione sembrano permettere che si crei la catena argomentale pro-associato, con la con- seguente capacita di questo di controllare dentro l’avverbiale, dimostrerebbero che l’associato a LF sale efjettivamente in Infl se si adempie il meccanismo rego- lare del passivo. La predizione che consegue da questa analisi é che nelle forme con gerundio co- ordinato I'associato debba essere unico controliore ammesso, data Pimpossibilita del controllo agentivo vista sopra (ess. 10a-13a, 14b-15b): il pro sarebbe ovviamen- te escluso come controllore, ¢ il suffisso passivo, come sappiamo, dovrebbe poier controllare solo nel gerundio di predicato. I dati in proposito non sono perd del tutto chiari. V. (23c)-(24c) con awverbiale di predicato, e (23d-e)-(24d-e) con avverbiale coordinato: con la sua mole (23) ¢. pro furono finite di costruire le nostre due case PRO lavorando ininterrot- tamente 4. *pro furono finite di costruire le nostre due case, PRO ricordandole poi come un lavoro immane €. pro furono finite di costruire le nostre due case, PRO avendo superato le difficolta iniziali (24) ¢. pro fu iniziata a costruite una chiesa PRO lottando con mille difficolta tec- niche 4. pro fu iniziata a costruire una chiesa, PRO ricordando poi la cosa come ‘un’ impresa estenuante ¢. pro fu iniziata a costruire una chiesa, PRO avendo trovato i fondi 15Lidia Lonzi Forse non deve stupire - data la natura limite della stessa costruzione in esame - che non siano possibili giudizi decisi, tanto pitt cbe il controllo di frase resta predi- cibilmente accessibile, come ogmino pud verificare, e questo pud forse essere all’origine della non totale inaccettabilita del controflo arb, almeno per qualche par- lante, in (23d-e), (24d-e), 8 Aspetti informativi della costruzione espletiva Per quanto ho sostenuto fin qui, se nelle frasi passive I’associato non controlla ¢ perché non sale in Infl, non avendo caso nominative. Nella costruzione passiva di tipo “transitivo” qui studiata, Vassociato, come abbiamo visto, & interpretato come un oggetto, e il PRO seleziona il pro soggetto come antecedent c-comandante. Lrimpossibiliti che si osserva per il controilo dell’associato in (25) & quella che si osserva per l’oggetto nella frase transitiva in (26): (25) *pro é stato investito Giovanni; PRO; correndo (26) *pro hanno investito Giovanni; PRO; correndo Che cosa ci sarebbe, perd, all’orine di questa sorta di bloceo del meccanismo del passivo ¢ della conseguente sahta lell’associato? é sufficiente pensare, come bo suggerito, a un effetto generato dalla presenza del pro espletivo che pud fungere da soggetto? Una parziale motivazzone del fenomeno potrebbe essere trovata sul piano della struttura informativa: non ¢*é salita a LF perché la frase non é predicativa. La strut- tura dell’espletiva ¢ una struttura eventiva in cui l'associato non ¢ il soggetto della predicazione: o meglio, non ”é predicazione La promozione del pro ad argomento esterno qui suggerita é solo un fenomeno di natura interpretativa che “riabilita” un espletivo asseynandogiii tratti propri dell impersonale. La frase con soggetto aggiunto al predicato (soggetto invertito), invece, ¢ ricon- ducibile a una frase predicativa, con una struttura Tema-Rema che abbiamo ragione di pensare valga anche a LF attraverso la regola di Focus (v. Chomsky, 1981). Ab- biamo prove che questo tipo di soggetto controlla come qualunque soggctto prever- bale (v. nota 1), Se osserviamo sotto questo profilo I’espletiva passiva a ristrutturazione, mi sem- bra che lo status dell ‘associato si avvicini a quello del sogyctto della predicazione: & infatti rappresentato in maniera naturale da un DP piuttosto che da un NP (nel senso di Chomsky, 1995, 342), e questo potrebbe derivare dalla stessa attuazione del mec- canismo del passive ma anche dal fatto che i verbi aspetuali passivizzabili nel pas- sivo lungo danno valore presupposto al verbo complemento (sappiamo per es. che se si asserisce che un processo é finito si tratta di un processo presupposto). 16Hcontrollo argomentale nelle costruzioni espletive passive Penso che vada comunque evidenziata l'apparente contraddizione che consegue da questa ipotesi. Dato che anche le espletive inaccusative hanno valore eventivo, perché in questo tipo di frasi dovremmo invece assumere la salita a LF del!’associato? La risposta owvia é che in queste frasi l’associato non puo che rice- vere Caso nominativo e, pertanto, salire in Infl per verificare i propri tratt, in quanto la struttura inaccusativa non permette che s"jnneschino altri meccanismi. 9 Conclusioni finali La predizione secondo cui un associato chiaramente nominativo deve poter con- trollare all"interno dell’avverbiale di un’espletiva passiva, @ confermata dalle frasi con forma pronominale nominativa. Anche se queste frasi presentano una struttura informativa particolate, e sarebbero forse da ricondurre a casi di soggetto invertito, Posservazione del predicibile fenomeno concomitante, cioé il mancato controllo del pro all’interno del gerundio coordinato, permette di concludere che il Caso nomina- tivo determina jt controllo in manicra univoca, quatunque sia l’analisi adotiata per la struttura in questione. Inoltre, abbiamo visto che nel passivo a ristrutturazione I’associato pud control- lare, ¢ ho suggerito che questo dato, pur non del tutto chiaro a un'attenta verifica delle sue predicibili implicazioni, potrebbe essere spiegato con l’attuazione obbliga- ta del meccanismo del passivo, per cui il DP pertinente pud fare checking di Caso solo in Infl. Pertanto, 11 fatto che il controllo argomentale sia marginalmente acces- sibile nel passivo lungo, potrebbe essere una prova a favore dell’analisi su cui ora basata la stessa aspettativa del controllo argomentale nell’espletiva passiva. Ho cercato infine di indagare se sia sostenibile che la mancata salita dell’associato nell’espletiva passiva, e la conseguente mancata correlazione tra ac- cordo € controllo, siano effetto di un’interazione tra struttura configurazionale € struttura informativa. L'idea che Massociato non controlli perché la frase espletiva & una frase eventiva, porta subito a una contraddizione apparente: la stessa ragione dovrebbe valere anche nel caso dell'espletiva inaccusativa, dove, invece, il controllo dell’associato & accertato, L’inevitabile via d’uscita ¢ che nell’espletiva inaccusativa Tassociato pud verificare il Caso solo in Infl. mentre nella passiva l’associato puo surrettiziamente Verificare il Caso anche in zi... grazie alla bivalenza del participio passato, Nel passive lungo, in cui il caso nominativo dell’associato sembrerebbe in- dipendentemente gurantito dal regolare meccanismo del passivo, Vassociato sembra anche costituire, per un insieme di ragioni da approfondire, il soggetto della predi- cazione e non il semplice oggetto di un cambiamento. 17Lidia Lonzi Bibliografia Baker, M., K. Johnson, & I. Roberts. 1989 Passive Arguments raised. Linguistic Inquiry. 20, 219-252 Belletti, A 1982 Morphological” Passive and Pro-Drop: the Impersonal Construction in Ttalian, Jou nut of Linguistic Research. 2, 1-34. Burzio, L. 1986. Italian Syntax, Foris, Dordrecht. Cardinaletti, A. 1995. Le posizioni soggetto dell italiano, XX] Incontro di Grammatica Generativa. Cardinaletti, A. 1997 Agreement and Control in Expletive constructions. Linguistic Inquiry, 28, 521-533 Chomsky, N. 1995. The Minimalist Program, MIT Press, Cambridge, Mass. Cinque, G. 1988. On “Si* Constructions and the Theory of “Arb”, Linguistic Inquiry. 19, 521-581. Cingue, G. 1997. Adverbs and Functional Heads. A Cross-linguistie Perspective, OUP, New York ~ Oxford. Cinque, G. 1998. The Interaction of Passive, Causative, and “Restructuring” in Romance. Round Table, 3-4 giugno, Universita di Padova. Dobrovie-Sorin, © 1998 Impersonal se Constructions in Romance and the Passivization of Uner- gatives, Litssuistic Inquiry, 29, 399-437. Giorgi, A. & F Pranest 1997.Tense and Aspect. From: Semantics to Morphosyntar. OUP, New York - Oxford Guasti, T. & L. Rizzi 1998 Non-agrecing do in child English and the expression of Agr. GLOW newsletter, 1998 Heltoft, L. & L. Falster-Jakobsen. 1995, Danish Passives and Subject Positions as a Mood System ~2 Content Analysis. ROLIG papir 54. Roskilde Universitetscenter. Lasnik, H. 1988, Subjects and the Theta Criterion. Natural Language and Linguistic Theory, 6, \-17. Lonzi, L. 1991. Frasi subordinate al gerundio, in L. Renzi eG. Salvi (a cura di), Grande Gramma- tica laliana di Consultazione, il Mulino, Bologna. Lonzi, L. 1998. Avverbi e altre costruzioni a controllo. il Mulino, Bologna. Pollock, J.-Y. 1981. On case and impersonal constructions, in R. May e J. Koster (eds), Levels of ‘Syntactic Representation, Foris, Dordrecht. 18ON TRUE DOUBLE OBJECT CONSTRUCTIONS IN TSHILUBA Gloria COCCHI J. Introduction In this paper I am going to examine Double object constructions (DOCs hence- forward) in Tshiluba, a Bantu language spoken in South-east Zaire. DOCs are in- diced typical of the whole Bantu family: in these languages, in fact, two preposi- tionless DPs may follow a ditransitive or complex verb. But differently from what happens in most Bantu languages (Swahili, Chichewa or Chimwiini, to mention just the most widely studied), both internal arguments of a Tshiluba DOC show a sym- metrical behaviour: both DPs can in fact be sentence initial and trigger subject agreement in a passive sentence, and V? may show optional object agreement in- fixes with either of them, and even with both at the same time, An example of a DOC in Tshiluba is given in (1), with the prototypical ditransitive verb, kupa ‘give’; we can see the peculiar behaviour of the two DP-objects as regards passivization ((2)) and optional affix-doubling ((3)): (1) mukaji u-p-a —muana_—_tshimuma woman st-give boy fruit “the woman gives the boy fruit” (2) a, muana u-p-ibu-a tshimuma —_kudi mukaji boy —surgive-pass__ fruit by woman “the boy is given fruit by the woman’ . tshimuma muana— kudi mukaji fruit su-give-pass boy by woman, “The fruit is given the boy by the woman’ 3) a. mukaji umu-p-a—muana_—_tshimuma woman su-io-give boy fruit 19Gloria Cocchi b. mukaji muana — tshimuma woman su-do-give boy frit ¢. mukaji u-tshi-mu-p-a mana tshimuma woman su-do-io-give boy fruit “the woman gives the boy frui The contrast is evident with a language like Swahili in (4)-(5), where only the indirect object displays such properties, while the direct object exhibits an inert be- haviour with respect to passivization and affix-doubling: (4) a. mtoto a-na-m-p-al — mwanamke —matunda boy su-tns-io-give woman fruit b. *mtoto a-na-ya-p-a - mwanamke —matunda boy —su-tns-do-give woman fruit ©. *mtoto a-na-ya-m-p-a - mwanamke —matunda boy su-tns-do-io-give woman fruit “the boy gives the woman fruit’ (5) a. mwanamke —a-na-p-ew-a matunda na mtoto woman su-tns-give-pass fruit by boy ‘the woman is given fruit by the boy’ b. #matunda ya-na-(m)-p-ew-a mwanamke na mioto fruit su-tns-(jo)-give-pass. woman —_by boy “the fruit is given to the woman by the boy’ The deep contrast between the two groups of Bantu languages as regards object properties of the two internal arguments in a DOC is weil known and has been widely studied in different times and different frameworks, from purely descriptive analyses to Relational Grammar, Lexical-Functional Grammar and Generative Syn- tax.2 The aim of this paper is to analyse Tshiluba DOCs by means of an alternative theory, stemmed from Minimalist program (as delineated in Chomsky (1993), (1995) and subsequent works), but substantially different from it in that Chomsky’s VP-shells are substituted by Clitic shells (as in Manzini and Savoia (1997), (1998), Savoia and Manzini (in prep.)). Not accidentally, VP-shells have been first proposed 1 Ditterently from Tshiluba, in Swahili the affix doubling the (human) indirect object (-m-) 1s mdeed obligatory, or at least strongly preferred. 2 See for instance Gary and Keenan (1977), Perlmutter and Postal (1984), Marantz (1984, 1993), Baker (1988, 1995), Kiparsky (1988), Bresnan and Moshi (1990), Cocchi (1991, 1992), Collins (1997), Nakamura (1997). 20On true double object constructions in Tshiluba mm order to analyse a particular DOC, English dative shift (see Kayne (1984), Larson (1988)). Notwithstanding, I will show that a VP-shell based system does not offer a satisfactory account of the Bantu data (especially as concerns Tshiluba), while a framework which employs Clitic shells seems to be more promising. 1 will first present some further Tshiluba data which confirm that the symmetri- cal behaviour of the two internal arguments of a ditransitive verb extends to other constructions, like applicatives and causatives.3 Then I will review some of the proposals put forward in the generative framework in order to account for dlitransitives in general. | will show how the Tshiluba data raise some problems for most theories on DOCs, and I will partly modify the current hypotheses in an attempt to account for these peculiar data, 2. Further data 2.1, Applicative constructions Similar to ditransitives is the case of applicatives. These constructions, typical of Bantu languages, involve complex verbs composed of a verb stem plus an applica- tive affix which introduces a new internal argument. This second object, which is generally called applied object for brevity, may have different semantic roles bene- factive/malefactive, goal/recipient, instrumental or locative.4 In many Tanguay ike Tshiluba, it is also possible to have a parallel construction involving a morpho- logically simple verb selecting a DP-do and a PP-io.5 It is interesting to notice that the order of the two internal arguments of a transi- tive applicative verb (the theme and the applied object) is the opposite with respect 3. Also transitive verbs with a locative complement may be considered as DOCS, since the locative-DP can passivize and trigger affix-doubling, as well as the theme-DP. Due to space problems I will not address this issue here. 4 In the examples I will gloss applied objects with io, due to their similarity of behaviour with indirect objects and abstracting from different semantic roles. Actually, only a sub- class of locative complements may be expressed by an applicative, namely those which Baker (1988: 244) calls ‘inner’ locatives. Cf. the contrast in (i) and (ii) (i) Islept in the bed The bed was slept in (inner locative) (ii) I slept in New York * New York was slept in (outer locative). 5 Languages vary on this point. Some languages like Kichaga do not have independent (ap- plicative) prepositions at all, while other languages like Chimwiini or Kinyarwanda have them for some roles (instrumental) and not for others (benefactive). Cf. Nakamura (1997), 21Gloria Cocchi to the order exhibited when the applied object is a PP, as in (6) below. This differ- ence pattems exactly with the altenance we find in English ditransitive verbs be- tween the non-shifted and the shifted sentence, as seen in the translations: (6) a. mukajiu-sumb-il-a muana_—_tshimuma woman su-buy-appl_—_ boy fruit “the woman buys the boy fruit’ b. mukajiu-sumb-atshimuma bua muana woman su-buy fruit for boy ‘the woman buys fruit for the boy’ As we can already expect, both internal arguments of an applicative clause like (6a) behave symmetrically, similarly 10 what we saw for ditransitives. Both can be passivized, as in (7), and the verb may exhibit optional affix-doubling with either (and both), as in (8). Furthermore, when one of the objects is passivized, the other can still show affix-doubling, as in ( (7). muana u-sumb-d-ibu-a? —tshimuma_—_kudi mukaji boy su-buy-appl-pass fruit by woman “the boy is bought fruit by the woman’ b. tshimuma tshi-sumb-id-ibu-a— muana_—_kudi mukaji fruit su-buy-appl-pass boy by woman “The fruit is bought for the boy by the woman’ (8) a. mukaji_—u-mu-sumb-il-a muana — tshimuma woman — su-io-buy-appl boy fruit b, mukaji —u-tshi-sumb-il-a muana_—_tshimuma woman — su-do-buy-appl boy fruit c. mukajiu-tshi-mu-sumb-il-a_muana —_tshimuma woman su-do-io-buy-appl_ boy fruit “the woman buys the boy fruit” © Bresnan and Moshi (1990), who analyse applicative constructions in another symmetrical language, Kichaga, evince other properties which characterize symmetrical languages like unspecified object deletion (deletion of patient in presence of another object) and re- ciprocalization (possibility to reciprocalize the patient, a5 well as the beneficiary). Due to space limitations T will not discuss these issucs here. 7 The applicative su(fix {-iF-] changes into [-id-] when followed by a palatal vowel. 22On true double object constructions in Tshiluba (9) a, muana wtshi-sumb-id-ibua tshimuma — kudi mukaji boy su-do-buy-appl-pass fruit by woman ‘the boy is bought fruit by the woman? b. tshimuma tshi-mu-sumb-id-ibu-a muana_—_kudi mukaji fruit surio-buy-appl-pass. boy by woman “The fruit is bought for the boy by the woman’ The situation is, again, different in Swahili, where only the applied object can be passivized ((10a)) and (must be) affix-doubled ((11a)). Expectedly, when the ap- plied object is passivized the patient cannot be affix-doubled, as in (12) (10) a. mtoto a-na-m-nunu-li-a mwanamke matunda boy su-tns-io-buy-appl_ woman fruit b. *mtoto a-na-ya-nunu-li-a mwanamke —matunda boy su-tns-do-buy-app! woman fruit c. *mtoto a-na-ya-m-nunu-li-a mwanamke —matunda boy st-tns-do-io-buy-app]_ woman fruit “the boy buys the woman fruit” (11) a. mwanamke — a-na-nunw-ti-w-a matunda na mtoto woman su-tns-buy-appl-pass fruit by boy “the woman is bought fruit by the boy’ b, *matunda ya-na-m-nunu-li-w-a mwanamke na mtoto fruit su-tns-io-buy-appl-pass woman by boy “the fruit is bought for the woman by the boy’ (12) *mwanamke —a-na-ya-nunu-li-w-a matunda na mtoto woman su-tns-do-buy-appl-pass fruit by boy ‘the woman is bought fruit by the boy” To complicate the matter further, in Tshiluba we can have an applicative con- struction out of an unaccusative verb, namely a verb which normally does not select, any object. The new argument introduced by the applicative, namely the applied ob- ject, behaves also in this case as a real object of the unaccusative verb: it can passiv- ize and trigger optional affix-doubling, as shown in the following examples: (13) a, muana w-lw-il-a mfumu boy su-come-appl_ chief b. muana u-mu-tu-ila mfumu boy swio-come-appl chief ‘the boy comes for the chie?Gloria Cocchi (14) mfumu wlu-id-ibv-a (kudi muana) chief su-come-appl.pass (by boy) “the chief is come for (by the boy)’ In symmetrical langage. on the contrary, not only it is not possible to have sen- tences corresponding to (136) o (14), but even (13a) is ill-formed. T exemplify this fact with Chichewa sentences (from Baker 1988: 255) (15) a. chiphadzuwa ——chi-a-fik-a beautiful woman su-tns-arrive “the beautifull woman has arrived’ b. *chiphadzuwa — chi-a-fik-ira mfume beautiful woman su-tns-arrive-appl chief ‘the beautiful woman has arrived for the chief?” 2.2. Causative constructions Differently from what happens in many languages, where causative construc- tions generally make use of a separate verb form (cf. English make/get, Italian fare, ete.), Bantu languages employ a verbal suffix, placed after the root like the applica- tive suffix seen above. Also in this case, the morphosyntactic operation has the ef- fect of introducing an extra argument with respect to the number of arguments se- lected by the base verb: the causer, namely the person/entity which causes the ac- tion encoded in the main verb to take place. When the main verb is unaccusative or unergative, the complex causative verb has only one mternal argument (the sole argument of the base verb, semantically the causee) and both symmetrical and asymmetrical languages behave alike: such DP can be passivized and affix-doubled (see Baker (1988) for an exhaustive discussion, and Cocchi (1992) for Tshiluba in particular). But when the base verb is transitive, the behaviour of the two objects reflects what we saw for applicatives and ditransi- tives: in languages like Tshiluba both objects behave symmetrically, except for word order (the causee must precede the theme) (exs, (16)-(18)), while in languages like Swahili only the causee, and not the theme, can be passivized and affix-doubled (exs. (19)-(20)): (16) a. mukaji u-sumb-ish-a muana_—tshimuma woman su-buy-caus boy fruit “the woman makes the boy buy fruit” b, *mukaji u-sumb-ish-a tshimuma muana 24On true double object constructions in Tshiluba (17) a, muana wsumb-ish-ibu-a —tshimuma —_kudi mukaji boy st-buy-caus-pass fruit by woman “the boy is made to buy fruit by the woman’ b. tshimume tshi-sumb-ish-ibu-a muana _—_kudi mukaji fruit su-buy-caus-pass boy by woman ‘*The fruit is made to buy to the boy by the woman’ (18) a mukaji— u-mu-sumb-ish-a muana —tshimuma woman — su-io-buy-caus8 boy fruit b. mukaji wtshi-sumb-ish-a - muana_—_tshimuma woman su-do-buy-caus boy fruit c. mukaji u-tshi-mu-sumb-ish-a_muana_—_tshimuma woman su-do-io-buy-caus boy —sfruit, ‘the woman makes the boy buy fruit” (19) a. mwanamke —a-na-m-nunu-ish-a foto. matunda woman su-tns-io-buy-caus boy ‘fut b. *mwanamke a-na-ya-nunu-ish-a toto. matunda woman su-tns-do-buy-caus boy fruit ©. *mwanamke a-na-ya-m-nunu-ish-a_mtoto —_-matunda wotnan su-tns-do-io-buy-caus boy fruit “the woman makes the boy buy fruit’ (20) a, mtoto —ana-numu-ish-ew-a —matunda_(na mwanamke) boy su-tns-buy-caus-pass fruit (by woman) “the boy is made to buy fruit (by the woman)’ b.* matunda ya-na-nunu-ish-ew-a mtoto (na mwanamke) fruit su-tns-buy-caus-pass boy (by woman) 8 Again, I will gloss with fo also the causee, namely the argument different from the theme, which behaves like an indirect object in asymmetrical languages. In the last section, how- ever, I will introduce a different terminology, and all these arguments (recipient, benefac- tive, locative, causee, etc.) will be grouped under the more appropriate label of Delimiter (from Borer (1994) and subsequent works), in that all of them share the aspec- tual/thematic property of delimiting the action expressed by the verb stem. 25Gloria Cocchi 3. Some reflections on ditransitive clauses and VP-shells In the earliest GB-literature, the structure of ditransitive verbs like give, which obligatorily select two internal arguments, a direct and an indirect object (respec- tively DO and 10 henceforward), made use of tripartite nodes: V’ dominated V° and its two arguments, DP-do and DP/PP-io. But since Kayne (1984), the theory allows only binary branching nodes; therefore another solution was needed for verbs con- taining more than one internal argument. Kayne (1984) proposes that 2 ditransitive verb has a complex structure consist- ing in two distinct clauses, the second of which has a locative meaning. Therefore a verb like give, instead of selecting a DP-object (like normal transitive verbs), selects another clause, headed by a (phonologically null) copular element, into which the P° heading the PP-indirect object incorporates when dative shift applies, and this is the reason why the indirect object surfaces as a prepositionless DP. The alternation be- tween the non-shifted sentence John gives a book to Mary) and the shifted one (ohn gives Mary a book), thus, parallels the alternation we find in loca~ tive/possessive clauses, along the lines traced by Benveniste (1966), who first ar gued that the incorporation of P? into BE yields HAVE, as in the well-known Latin alternance miki est filius vs. habeo filium. The structure of the two mentioned Eng- lish ditransitive sentences will thus look as follows: (21) a. John gives [ a book (is) to Mary] 'b. John gives [Mary (is+P°= has) a book] Larson (1988) develops Kayne’s suggestions and postulates that a ditransitive verb consists of two ‘VP-shells’: the complement of the highest V° (give) is indeed a second VP, headed by a phonetically empty verb (which incorporates in the high- est one in the course of the derivation), in whose Specifier and Complement posi- tions the two internal arguments are generated (22) [w I [ver su Vi [ver do V2 [ve P io ]]]] According to Larson, in order to derive from (22) the corresponding shifted sen- tence, we must assume a movement of the NP-io (intemal to PP) to Spec(VP2). namely the position occupied in (22) by the direct object; such a movement is made available by means of P? incorporation into V°, as in Kayne (1984). In Larson’s analysis, this kind of movement patterns with the derivation of passive sentences, where the direct object moves to subject position, and the subject becomes a ‘chomeur’ (in Perlmutter and Postal’s (1984) terminology, used by Larson): it loses subject properties and has to surface obliquely (or may not surface at all). Analo- gously, in the case at hand, the IO moves to the ‘subject’ position of the embedded 26On true double object constructions in Tshiluba Vp, where the DO is generated, and the latter becomes a “chomeur’: it loses object properties and surfaces in a right-adjoined position: (23) Lie T [ver su VI+V2 [vez [ver io; fv ty ] do II This analysis was proposed in order to account for the fact that in (22) the for- mer indirect object is the NP which shows all the properties of objecthood, while the direct object seems to have lost them. This is shown very clearly by the fact that, while in a passive non-shifted sentence the DO can be promoted to subject position, as in (24), in a shifted sentence this is not possible anymore: only the former IO can move to the subject position, as in (25): (24) John gives the book to Mary ‘The book is given to Mary *Mary is given the book to (25) John gives Mary the book Mary is given the book 2*The book 1s given Mary Even if Larson's theory has been subject to many changes and reformulations in the past years, it has been generally assumed since then that syntactically complex verbs like ditransitives are composed of (at least) two VP-shells, in order to comply with Kayne’s binary branching requirement. 3.1. VP-shells in the Minimalist Program The use of VP-shells has been recently extended by Chomsky (1995), He as- sumes in fact that even simple transitive verbs are composed of two VP-shells. In order to motivate such a structure, Chomsky assumes that the configuration v-VP represents the semantic meaning of ‘Cause’, crucially involved in all transitive sentences. Differently from Larson's proposal (cf, (22) above), Chomsky argues that the lexical verb is generated in the most embedded VP, while the higher shell is headed by a so-called ‘light verb’ (v”), which is a functional head. The structure of a transitive verb will thus look as follows (26) [re T° [vp su v° [vp V° dol) Chomsky analyses unergative verbs as hidden transitives, along the lines traced by Hale and Keyser (1991), Laka (1993) and subsequent works. Thus their structure will be the same as (26) for transitives. Unaccusative verbs will therefore be the only verbs consisting of a simple VP, as in (27): (27) [re T° [ve V° DPJ] 27Gloria Cocchi Chomsky (1995) does not make any reference to the structure of ditransitive verbs in a VP-shell based system, Bobaljik (1995) instead, following the previous version of the Minimalist framework (Chomsky (1993), which still makes use of AGR-heads, assumes that the structure of (shifted) ditransitive verbs consists of three VP-shells, each projecting an AGRP, as in (28) below. In this model, there- fore, the number of VP-shells - and agreement heads - strictly parallels the number of arguments selected by a verb. (28) Ere T Lacase AGRS [vp SUV [acaiop AGRio [yp 10 V [,cndor AGRdo [ve DO V]IIIII} In this structure, each DP checks Case in the immediately higher Spec(AGRP) position. In this way it is possible to avoid Equidistance problems and crossing paths when the DPs move from base position to Case-checking position, especially in the case of two internal arguments: the Case-checking position of the DO is lower than the base-position of the IO, and the same holds between IO and SU. ‘Once AGR-heads are dispensed with, in line with Chomsky (1995), the structure in (28) could be rewritten as in (29), which maintains the symmetry between the number of arguments and VP-shells. This is indeed the structure adopted by Collins (1997): (29) [rm T [ver SU v [2 10 v [ve V Do JI] In absence of AGR-heads, Chomsky (1995) assumes that the internal argument of a transitive structure, like (26) above, checks Casc against v°, and a second speci- fier is projected to accommodate the DP when it checks Case in overt syntax (since the DP-subject, or its copy, already occupies Spec(vP)). For a ditransitive structure like (29), Collins (1997) assumes that the DO checks Case ayainst v°2 and the 10 against v°1, by projecting a second specifier for each v°-head if necessary. Thus, the Case-checking modality strictly reflects Bobaljik’s hypothesis. 3.2. Problems raised by the Tshiluba data On the basis of Tshiluba data in (1-3) above, it becomes evident that the struc- tures of ditransitives hypothesized in (28) or (29) need re-discussing. We see in fact that both structures are not adequate to capture the situation of symmetrical languages like Tshiluba. These structures, in fact, can only account for DOCs in languages like Swahili, as well as for English Dative shifi, namely contexts in which the two objects show an asymmetrical behaviour, and the movement of the 9 1 am using Chomsky’s ter logy. Collins, instead, calls v1 and v2 respectively Tr* (from Transitivity) and Appl? (from Applicative). Furthermore, Collins generates the DO in VP-complement position instead of Spec(VP). This divergence is immaterial for the resent purposes. 28On true double object constructions in Tshiluba DO to the subject position of a ditransitive passive sentence must be ruled out (see (5b) above). ‘According to Chomsky (1995), a passive sentence, like an unaccusative one, should lack the highest vP-shell (as well as the eventual AGRsP), As a consequence T°, under Minimal Link Condition, attracts the higher-generated DP, the indirect object, which moves to Spec(TP) and checks nominative Case, while the direct ob- ject checks accusative Case by moving (presumably at LF) to Spec(vP2) / Spec(AGRdoP). The impossibility for the DO to move to Spec(TP) is thus captured by Minimality/MLC teasons: in both (28) and (29) the 10, a potential checker for the D°-feature in T°, is nearer to Spec(TP) than DO (cf. Rizzi (1990)). But this ex- planation does not account for Tshiluba, where the DO can move to subject position even in presence of an 10. Furthermore, if the asymmetries between IO and DO in Swahili were entirely due to minimality considerations, we could account for the passivization facts but not for the affix-doubling facts, and it is quite unlikely that two so strictly related phenomena should receive completely separate accounts 4. A tentative proposal: parametrization of multiple Case feature-checking In an attempt to overcome the problems raised by the Tshiluba data, we may as- sume that the structure of ditransitives consists in only two VP-shells; the structure with three VP-shells, in fact, does not prove to be universally valid. Thus, the structure of ditransitives should pattern with that of transitives (cf. (26) above): the DP-subject is merged in the specifier of the highest vP, while the two objects are both generated VP-internally, in the specifier and complement position of the lowest shell, a hypothesis which recalls Kayne’s and Larson’s proposals dis- cussed above. In this way the two objects are equidistant from subject position, Spec(TP), once the DP-subject is not merged (in passives). The various languages establish, as a means of parameter setting, the relative position of the two objects, as well as whether they are both prepositionless DPs, as in Bantu, or a DP and a PP, as in Romance.10 The structure of both Tshiluba and Swahili will thus look as in (30): (30) [re T [ve SU v [vp 10 V_ Dojf} 10 Jn Romance languages, for instance, the DO will be merged in Spec(VP) and the PP-1O in the Complement. In this case the two DPs are not equidistant from Spec(TP), due to the PP-node, and this accounts for the fact that only the DO can passivize. A language like English allows instead for two possibilities, arising from two different numerauions one with a PP-IO in the Complement position, like in Romance, and one with a DP-I0 in the Specifier (the so-called shifted sentence), but with an inert DO, like in Swahili 29Gloria Cocchi In absence of the subject-DP, both objects are equidistant from Spec(TP), thus either of them can move to such a position and become the surface subject of the sentence, triggering subject agreement. This is consistent with the Tshiluba data in (2) above, but we must still explain why the movement of the DO is not possible in Swahili ditransitives; in other words, something must block the movement of the DO in asymmetrical languages! ! The contrast between Tshiluba and Swahili could be captured by a parameter concerning the status of the DO in a DOC (as was in Baker (1988), Cocchi (1992)): we can assume that in Tshiluba, but not in Swahili, also the DO is a real object, structurally Case-marked.!2 In Minimalist terms, the parameter relates to the possibility for the accusative Case feature contained in a ditransitive or com- plex V°!3 to erase afier being checked once (by the 10), thus admitting (or not) multiple checking. * We can assume that, in languages like Tshiluba, multiple checking is allowed; therefore the derivation of an active ditransitive proceeds as in (31): (31) [yp SUT [ve S¥[,) f£(D0)-fi(10)-v [ye 10 V DoJ})) In passives, either of the objects can moye to Spec(TP), while the other still checks Accusative Case against v°. Here | assume, following Collins (1997)!4 and 11 To assume a structure with two VP-shells for Tshiluba and one with three VP-shells for Swahili (for the same construction) should go against any principle of Economy, and the- refore we reject such a possibility. 12 This is in fact Baker's (1988) assumption for asymmetrical languages like Swahili, He proposed the Case Frame Preservation Principle (CFPP), which says that a complex verb cannot assign more Cases than a simple one. Therefore the [O receives the sole structural Case, while the DO receives inherent Case or is visible through noun reanalysis into in both cases it has an inert behaviour. As already noted by Cocchi (1992), such a prin ple obviously does not hold for Tshiluba 13 Crucially, not every transitive verb in Tshiluba admits multiple objects, but only those verbs which are morphologically complex, namely resulting from the incorporation into 'V° (in the lexicon) of a Case-assigning element (P° in ditransitives and applicatives, V° in causatives). In other words, the complex verb seems to ‘inherit’ from its intemal constitu- ents the possibility of assigning one Case each. See at this regard the proposals in Ma- rantz (1984) and Cocchi (1992), who argue against Baker's Case Frame Preservation Principle. 14 Collins (1997) assumes that vP is present also in unaccusatives, and hence in normal pas- sives. He motivates such an assumption with data relative to word order in English loca- tive inversion clauses. In the case I am discussing, the presence of vP is further motivated by the need of checking Case on that object which does not check nominative. 30On true double object constructions in Tshiluba contra Chomsky (1995), that the highest vP, in whose specifier the DP-subject is merged, is present also in passives, though in these cases no DP is base-gencrated in its specifier. The derivation proceeds as in (32): (32) [re 10/D0 T [vr ff(DO/O)-v [vp (10) V_ (D0) TI] On the contrary, we can assume that in Swahili multiple accusative checking is not permitted, since the Case feature contained in V° erases after checking the Case of the IO. As a consequence, the DO is not a structural object (and does not have the related properties; cfr. (4)-(5)), and it is visible at PF thanks to other processes, like by receiving inherent Case (cf, the discussion in Baker 1988). Therefore, the sole DP which can move to Spec(TP) in passives is the IO, which still has an unchecked Case feature: 3) [ 10 T [w ¥ [vp #@ V_ DoT) The explanation that Chomsky (1995) offers for Multiple Subject Construc- tions in Icelandic is indeed similar: the (Nominative) Case feature contained in T° does not erase after being checked once. The same, I assume, happens in Tshiluba for the Accusative Case feature contained in V°. Not accidentally, both Multiple Subject Constructions and Symmetrical Double Object Constructions are not very widespread cross-linguistically; indeed they represent exceptions to the rule, as exceptional is the fact that an uninterpretable feature may enter multiple checking relations. 4.1. Residual problems To sum up, in the preceding section I have assumed that the structure of ditransi- tives does not consist of three but rather two VP-shells (contra Bobaljik (1995) and Collins (1997)) on the basis of the Tshiluba data, which a structure with three VP- shells 15 not able to account for. Furthermore, I have argued that the various lan- guages establish, as a matter of parameter setting, the relative position of the two internal arguments (both generated in the lowest VP), as well as the number of structural (accusative) Cases that V° can assign. Though this proposal might sound attractive, some residual problems remain open. First, in Swahili, although both DP-objects are equidistant from the accusative Case-checking position (Spec(vP)), as well as from Spec(TP), the data show that only the 10 exhibits object properties. We have assumed that the IO checks the sole Case available, while the DO gets inherently Case-marked. At this point we might wonder why it is necessarily the IO, and not the DO, which checks the sole Accusa- tive Case feature of the verb, being the two equidistant from Spec(vP) 31Gloria Cocchi Second, though a structure like (32) above can account for the passivization data in Tshiluba in terms of MLC, the connection between passivization and affix- doubling remains unexplained. Furthermore, Baker's assumption of affix-doubling as a diagnostic of true objecthood, namely as a proof of accusative Case marking, is highly speculative and objectable. If Bantu affixes can be considered as the bound counterpart of Romance clitics, in fact, it is evident that the latter do not necessarily double accusative objects. Third, the DPs which can passivize or be affix-doubled in Tshiluba can even be three in some complex cases, e.g. when a verb is causative and applicative at the same time, as in (34):15 (34) mukajiu-sumb-ish-il-a mfumu — muana_—_tshimuma woman su-buy-caus-appl chief boy —_—fruit “the woman makes the boy buy fruit for the chief” The problem we meet in this case is that, unless we allow ternary branching (as, did Baker (1988), there is no possible way to have three DPs which are equidistant from a certain position, Spee(TP) in this case. Last but not least, an unaccusative applicative sentence like (13a) above raises many questions once we analyse its structure more in detail. Since we have assumed (30) as the base structure of Tshijuba ditransitives, with the applied object in Spec(VP) and the theme in compiement position, the structure corresponding to (3a) should be (35), as the sole argument of an unaccusative is thematically a theme and not an agent: (35) Gre T Ge v [ve 10 V_ Dof]} The problem with this structure is that both DPs are equidistant from Spec(TP); thus T° could in principle attract either of them, to check its EPP feature, in the active sentence. But if the applied object moves, the sentence gets a different semantic inter- pretation: the theme, which remains in situ, is indeed understood as the applied object: (36) *mfumu —uelu-ila muana chief su-come-appl_ boy “the boy comes for the chief” OK as ‘the chief comes for the boy” 15 This was indeed Cocchi’s (1992) main criticism to Baker's (1988) CFPP. Baker said in tact that, since a simple verb (like kupa ‘give’ in (1) above) may assign two Cases in lan guages like Tshifuba, also a complex verb might do it, and this should account for the symmetrical behaviour in applicatives and causatives. But there are obviously no simple verbs wnich can assign three Cases. 32On true double object constructions in Tshiluba A possible way out could be to assume that while kulua is an unaccusative verb, the applied form kuluila is not, because the DO does not have theme proper- ties any longer, but rather agent ones, typically mental volition (cf. Reinhart (1996)). But this analysis is also far from being attractive; besides, it does not ex- plain why a sentence like (13a) is not allowed in languages like Chichewa ot Swahili (cf.(15) above) 5. An alternative theory: Clitic shells instead of VP-shells In the preceding section we have seen that a VP-shell approach to the structure of DOCS in languages like Tshiluba and Swahili leaves some fundamental questions un- answered. Therefore, it might’be worthwhile to explore other possibilities offered by current theories of syntax, still in the spirit of Chomsky’ (1995) Minimalist program. One such theories is presently proposed by Manzini and Savoia (1997, 1998), Savoia and Manzini (in prep.), and it is based on a detailed study of clitic construc- tions in Romance languages, with a particular emphasis on Italian dialects, As dis- cussed before, Bantu subject- and object-affixes present a lot of syntactic similari- ties to Romance clitics; therefore the assumptions formulated for the latter can prove extremely useful for an analysis of the former In their works, Manzini and Savoia substitute the VP-shell module with an ar- ticulated set of Clitic projections generated above I° (Clitic shells). Adapting origi- nal suggestions of Sportiche (1992), they hypothesize that each clitic heads a sepa- rate projection CIP. Since there are clitics which murually exclude each other (like, for instance, locative and dative clitics), we can reasonably argue that they realize the same CI°; therefore it is misleading to identify such CIPs in terms of Case: they are better characterized in terms of thematic/aspectual properties of the verb, in the sense of Borer (1994) and related works.16 In this model the Agent is seen as the Originator (Or) of the event, the Patien/Theme as the Event Measurer (Meas) and the Recipient/Goal, as well as the Locative, as the Delimiter (Del). Therefore, the positions for IO- and DO-clitics can be identified respectively as CiDel® and ClMeas®; the subject clitic, which is aspectually non-differentiated, can be simply indicated as CID°, In (37) below we can see the structure of the various projections assumed for Standard Italian and Italian dialects (from Manzini & Savoia (1997)): (37) CID® Ciel? CiMeas® 1°/T? ve 16 Sce Tenny (1992, 1994), Arad (1995, 1996), Cocchi (1998), Manzini and Savoia (1997, 1998). 33Gloria Cocchi ‘As a consequence, Manzini and Savoia assume that the arguments of a verb are not represented by the full DPs present in the sentence, but rather by the clitic pro- nouns which, obligatorily or optionally, are attached to the verb (preceding or fol- lowing it according to its finiteness). Thus, the full DPs only ‘double’ the clitics, similarly to what happens in the widely studied cases of clitic doubling structures in languages like Spanish (cfr. Sportiche (1992), Torrego (1994) and many others). In this system a clitic gets linked with the respective aspectual feeture/theta role (contained in V°) by attracting it with a covert operation of Attract-F.!7 Theta roles, thus, behave like the other pure features (contra Chomsky (1995), whe assumes that theta roles are not features). If we assume in fact, as in Platzack (1994) and Manzini and Savoia (1997), that an empty category like pro is not justifiable in a Minimalist framework, theta-marking of the clitic cannot be achieved with movement of a pro (generated in theta position) to Spec(CIP), and another solution is required. But if theta roles can be attracted like the other features, there is no reason to hypothesize VP- intemal generation of the full DPs either; Manzini and Savoia assume indeed that the DPs are generated directly in their spell-out positions. The sentence-initial DP (traditionally seen as the surface subject) is presumably in the Spec of a TopicP above CIDP (in null subject languages like Italian, or in clitic subject languages like Norther Italian dialects, as well as in Bantu languages), !8 while the internal argu- ments are merged in the specifier position of FocusP or TopicP lower than VP. 5.1. Clitie shells and Bantu languages The framework I have just delined sheds new light on the problem raised by the Bantu data, and especially by the contrast between symmetrical and asymmetrical languages, and opens new ways towards a satisfying account of the data The perspective of the analysis is indeed completely reverted from current ac- counts. The contrast between languages like Tshiluba and Swahili is in fact gener- ally analysed in terms of Case features, and/or in terms of locality/MLC (for what 17 Thus ClDel® and ClMeas® will attract respectively the Delimiter (recipient, locative, ap- plied object, cause, etc.) and the Mcasurer (patient, theme, partitive, etc.). CID®, which is aspectually non-differentiated, allows more possibilities: it will attract the Originator (agent, cause, etc.) in transitives and unergatives, but in the case of unaccusatives or pas- sives it may attract one of the other two roles (depending on the language in question), and it may even attract more than one, as in teflexives (e.g. Originator and Measuret). 18 In overt subject languages, like English, the DP-subject is instead generated in Spec(CID), as it must check the strong D-feature. In fact we cannot assume that such a fe- ature is weak in English, as the presence of the DP-subject is obligatory, and it cannot be checked by a clitic/afffx, as English does not have these elements. 34On true double object constructions in Tshiluba concerns passivization), and we have seen that several problems remain unexplained (of section 4. .). In the framework at hand, Case matters have very little - if any - weight, especially if we consider that clitics/affixes are treated as the real arguments of the verb (be they overt or covert), while full DPs are only doubling elements, thus swith no Case. Vice versa, locality plays no role at all, as the DP-surfuce subject of a passive sentence is merged directly in its spell-out position, Spec(TopicP). There- fore, the fact that some passive structures are ill-formed in Swahili, but not in Tshi- |uba, cannot be ascribed to constraints on movement. Manzini and Savoia assume that each clitic head contains features, for example a D®-feature,!9 which must be checked in syntax (if strong) or at LF (if weak). The sirength of these features are established by the Various languages by means. of pa- ruineter setting. Checking of a strong feature can be achieved by merging a special- ized clitic in the head itself, if the language under analysis has clitics (like Romance or Bantu languages), otherwise by merging or moving a DP into the specifier posi- tion of the clitic head in question (like Germanic languages) These assumptions could help account for the different distribution of affixes in Tshiluba and Swabili: for instance we could assume that the two languages pa- rametrize differently with respect to the strength of (some features of)CIMeas*: it could be strong in Tshiluba and weak in Swabili, thus accounting for the contrast between (3b) and (4b) above. But this is not the correct prediction either: in (38) we see in fact that a ClMeas? can be realized in Swahili, but only if a CIDel® is not present, namely in simple transitives: (38) mtoto a-na-ki-nunu-a —_(kitabu) boy — su-tns-do-buy (book) “the boy buys it (the book)’ There is however a possible way out to this impasse. Cross-linguistic data show indeed that postulating a universal structure like (37) above is too strong a hypothe- sis, for what concerns either the number or the relative order of clitic shells. We could therefore say that languages parametrize on these two points. For instance, the Tshiluba data in (39a) show that the relative order of ClDel® and ClMeas® is the op- posite of what we find in Italian (39b) and its dialects, or Spanish, but it reflects the French order (39c) (though the order of the clitics is completely rigid within the same language): 19 This D°-feature encodes categorial feature (all clitics are nominal), as well as definite- ness. It patterns with the D°-feature also contained in [°/T°. 35Gloria Cocchi 39) a. mukaji (CIMeas > C1Del) woman su-do-io-give b. 1a donna glie-lo da (CiDel > CiMeas) woman —io-do gives c. lafemme le lui donne (CiMeas > CiDel) woman do io gives. ‘the woman gives it to him’ Analogously, we could say that languages parametrize with respect to the num- ber of clitic arguments a verb selects. Recall that Manzini and Savoia (1998) assume that, when two clitics are mutually exclusive (like locative and dative in Italian), this is an evidence that they occupy the same thematic slot, namely that they realize the same clitic head.20 This amounts to saying that we can reasonably assume that, while languages ike Tshiluba project a ClIMeasP as well as a ClDeIP, as in (40) below, languages like Swahili project only one Ciitie shell for both aspectual roles, as in (41). This Del/Meas clitic head can alternatively be filled by a Meas-clitic, as in (38), or by a Del-clitic, as in (4a) or (10a) above, but of course not by both at the same time, and this accounts for the ill-formedness of (4c), in contrast with the well-formedness of (3c) in Tshiluba. (40) [ewe U [cimease tShi [cir mu [yp p-a [yp..... (ef. (30)) (41) [ower @ Ere na [epenmease M Lip pra [ype (of (4a)) The fact that, when the verb is ditransitive, only the Del-clitic can fill CIDel/Meas? in Swahili correlates with the traditional assumption that, in this case, only the Del/O is a real object of the verb,2! while the theme is a chomeur, a sec- ond object, etc. In our terms, since arguments are expressed by clitics, and the Meas/DO cannot be a clitic if a DeV/IO is selected, this amounts to saying that the theme is no argument at all. in Swahili ditransitives. 5.2. An account for the residual problems The assumptions made in the previous section can also account for the other questions which remained unexplained (cf. 4.1.). 20 This was indeed the main evidence for identifying clitic heads in terms of aspectual prop- erties rather than in terms of Case and/or in terms of traditional theta-roles, 21 Cf, Perlmutter and Postal (1984), Marantz (1984), Baker (1988), Bresnan and Moshi (1990), Cocchi (1992) and many others. 36On true double object constructions in Tshiluba Let us start with the lack of passivization of the DO in Swahili, which, I argue, can be made to correlate with the lack of a separate ClMeas® in this language. I assume in fact that, in passives, the position CID® is first unfilled, as there is no Onginator; afterwards one of the clitics, base-generated in a lower position (where it gets aspectually identified) moves to CID® (cf. fa.17). In Tshiluba two different affixes/arguments are available, while in Swahili only one, since there is only one clitic head below CID®. Such a head can be filled only by a Delimiter in dutransitives, therefore the subject of a ditransitive passive can only be a Delim- iter.22 Since the full sentence-initial DP is, in this view, only a correlate of the affix, and it is base-generated in Spec(TopicP) above CID®, the consequence is thar either of the two DP-objects can appear in sentence-initial position in Tshi- luba, but only the Delimiter/1O in Swahili, since the sentence-initial DP must agree in features with the affix in CID? Another problem was the fact that in Tshiluba we can have up to three DP- objects. In this view, this amounts to saying that in this language we can (exception- ally) have up to three CIPs below CID°, in marked situations. Therefore, in cases Ike (34) above we have three different clities which can fill CID° when the sentence 1s passive, and three different DPs which can be merged in Spec(TopicP). This, of course, is not possible in Swahili, where we can have only one CIP below C1D*: (cf. (34) Last but not least, there is the question of the existence of unaccusative applica tives in Tshiluba. T assume that in this case the two (clitic) arguments are base- generated respectively in ClMeas? and in C1Del®, where they get aspectually inter- preted. If CIMeas* moves to C1D® position (which is empty, being the verb unaccu- sative), the theme is understood as the subject (and can be doubled by a DP in sen- tence-initial position, Spec(TopicP)), while the other affix, generated in C1Del®, rep- resents the applied object, which can be doubled by a DP in a peripheral position lower than VP, as in (43). Of course, if it is the ClDel? which moves to C1D® in- stead, we derive the opposite interpretation, (43) [rope muana [cor U [emease & [era mu [rp bila [vp.... mfumu (cf. (13a)) If this sentence is made passive, the Del-affix climbs to CID®, which is now empty, and the applied object can be merged in Spec(TopicP). Once again, in Swa- hili we cannot have an applicative unaccusative (even in the active form) since ClMeas? and ClDel° are mutually exclusive (42) [ewe U [ermesse thi [ewer mu [ope mu [jp sumb-ish-il-a [vp 22 Of course, in simple transitives, the Meas/theme can passivize, but we have also seen that in these contexts it can also be affix-doubled, Once again the two properties go together. 37Gloria Cocchi Conclusions To sum up, in this paper I have analysed DOCs in two Bantu tanguages, Tshi- luba and Swahili, which do not behave alike with respect to passivization and aftix- doubly in Tshiluba both objects have a symmetrical behaviour, while in Swabili only th. mdirect object can passivize and be affix-doubled, T have first shown that most of the current proposals are able to account for the situation of Swahili but not for that of Tshiluba. Afterwards I have discussed several possible sentence structures for ditransitives in terms of a Minimalist, VP-shell- based framework, and [ have shown that all of them either do not account at all for symmetrical languages, or they leave some fundamental questions unanswered, Fi- nally T have discussed a Clinc shel!-based approach (from Savoia and Manzini (in prep.) and related works), which has proved adequate to overcome most of the prob- ems raised by the Tshiluba data and, furthermore, looks very promising in view of a common account of Bantu affixes and Romance clitics, References Arad, M. (195) ‘On the projection of ditransitive verbs’, UCL Working Papers 7: 215-233, Arad, M. (1996) “A minimalist view of the syntax - lexical semantics interface’, UCI. Work- ing Papers 8: 215-242. Baker, M. (1988) Incorporation, Chicago and London: The University of Chicago Press. Baker, M. (1995) The Polysynthesis Parameter, Oxford: Oxford University Press Benveniste, E. (1966) Problémes de Linguistique Générale, Paris: Gallimard. Bobaljik, J.D. (1995) Morphosyntax, Ph.D. dissertation, MIT. Borer, H. (1994) ‘The projection of arguments’, in: E. Benedicto and J, Runner (eds.) Func- tional Projections, University of Massachusetts Occasional Papers \7: 19-47, Bresnan, J. and L. Moshi (1990) ‘Object asymmetries in comparative Bantu syntax’, Linguis- tie Inguiry 21/2: 147-185. Chomsky, N. (1993) “A Minimalist Program for linguistic theory’, MIT Occasional Papers in Linguistics 1, Cambridge MA: MI T Press. Chomsky, N. (1995) The Minimalist Program, Cambridge MA: MIT Press. Cocchi, G. (1991) ‘I tre passivi del Tshiluba’, Quademi del Dipartimento di Linguistica 2, Universita di Firenze: 17-42. Cocchi, G. (1992) ‘Lingue che ammettono due veri accusativi: Il caso del Tshiluba’, Rivista di Grammatica Generativa 17: 101-140 Cocchi, G. (1998) ‘Ergativity in Romance languages . in: ©. Fullana and F, Roca (eds.) Stud- ies on the Syntax of Central Romance Lunguages, University of Girona: 83-99. Collins, C. (1997) Local Economy, Cambridge, MA: MIT Press. Gary, J. and E. Keenan (1977) ‘On collapsing grammatical relations in Universal Grammar’, P. Cole and J. Sadock (eds.) Syntax and Semantics 8: 83-120. 38On true double object constructions in Tshiluba Hale, K. and S. J. Keyser (1991) ‘On argument structure and the lexical expression of syntac- tic relations”, ms., MAT. Kayne, R.S. (1984) Connectedness and Binary Branching, Dordrecht: Fotis Kiparsky, P. (1988) ‘Agreement and linking theory’, ms., Stanford University, CA Laka, I. (1993) “Unergatives that assign ergative, unaccusatives that assign accusative’, im J.D. Bobaljik and C. Phillips (eds.) Papers on Case and Agreement, MIT Working Papers m Linguistics [8: 149-172 Larson, LK. (98S) *On the double object construction’, Linguistic Inguiry 19/3: 335-391 Manzini, M. Rand L M. Savoia (1997) ‘Null subjects without pro’, UCL Working Papers 9: 303-313. Manzini, M. R. and L. M. Savoia (1998), ‘Parameters of subject inflection in Italian dialects’, in; P. Svenonius (ed.) Proceedings of the Workshop on the EPP (Tromso, June 1997), New York Oxford University Press. Marantz, A (1984) On the Nature of Grammatical Relations, Cambridge MA: MIT Press. Marantz, A (1993) ‘Implications of asymmetries in double object constructions’, in: S. Mchombo (ed.) Theoretical Aspects of Bantu Grammar, Stanford CA: CSLI Publica- tions, 113-150. Nakamura, M. (1997) ‘Object extraction in Bantu applicatives: Some implications for Mini- malism’, Linguistic Inguiry 28/2: 252-280. Perlmutter, D.M, and P.M. Postal (1984) ‘The I-advancement exclusiveness law’ in: D.M. Perlmutter and C.G. Rosen (eds.), Studies in Relational Grammar, Vol. 2, Chicago and London: The University of Chicago Press, 81-125. Platzack, C. (1995) “Null subjects, weak Agr and syntactic differences in Scandinavian’, Working Papers in Scandinavian Syntax 53: 85-106. Reinhart, T. (1996) ‘Syntactic effects of lexical operations: Reflexives and unaccusatives', UiL OTS Working Paper, Utrecht University. Rizzi, L. (1990) Relativized Mintmality, Cambridge MA: MIT Press. Savoia, LM. and MLR. Manzini (in prep.) J Dialetti Ialiani, Bologna: I Mulino. Sportiche, D. (1992) ‘Clitie Constructions’, ms,, UCLA. Tenny, C. (1992) ‘The aspectual interface hypothesis’, in: I. Sag and A. Szabolesi (eds.) Lexi- cal Matters, Stanford CA: CSLI Lecture Notes, 1-27 Tenny, C. (1994) 4spectual Roles and the Syntax-Semantics Interface, Dordrecht: Kluwer. Torrego, E (194) ‘On the Nature of Clitic Doubling’, ms., UMASS. 39CONSTRAINING CHECKING: ON ©-FEATURES REPRESENTATION AND THEIR PARAMETRIC VARIATION* Elisa DI DOMENICO 1, Introduction Inflection and agreement are two interrelated processes in natural languages. While agreement can be characterized as the process by which a giammatical element X matches a grammatical element Y in property Z within some grammatical configuration, inflection concerns the assignment of agreement or other properties to individual lexical items. The advantage of inflection seems clear: it saves space in the lexicon in that it enables the introduction of new information concerning a lexical item maintaining the same stem. Less clear seems the function of agreement Jespersen (1922) defines it ‘superflu- ous’, although he admits that agreement might have the advantage of allowing word order to be freer, Haiman (1985) considers 1 “notoriously disfunctional’, as pointed out by Ferguson and Barlow (1988) who recognize, on the contrary, some important functions of agreement. First, multiple cues are better than a single cue when the Latter is attenuated, missing, or masked by other material, Second, agreement may license This paper reports the provisional results of a work in progress I am doing at the Univer- sity of Florence with a post-doctoral fellowship. 1 thank Maria Rita Manzini for her comments on this work. I also thank her, together with Leonardo Savoia for the stimulat- ing environment they have created at the Department of Linguistics in Florence and for sharing their unpublished data with me. Lowe special thanks to Adriana Belletti and Luigi Rizzi for widening and fruitful discussion, Finally, I thank Valentina Bianchi, Annita De Amicis, Richard Kayne, Giuseppe Longo- bardi, Cecilia Poletto, Ian Roberts All errors and omissions are of course my own. 41Elisa Di Domenico use of word order for topicalization, focusing and other functions, or it may license the omission of case marking on nouns in certain grammatical relatio Whether useful or useless, inflection and agreement are not universally attested. A wide group of languages, namely those traditionally defining the isolating type, lack both phenomena, as shown in (1): (1) a. You ye tido mad zai huadyudn- Ii Mandarin Chinese exist one CLASS. eat at garden in there is a cat in the garden b. You san-tiéo mad zai hudyudn- li exist three CLASS. cat at garden in there are three cats in thy garden In (1), the fact that the subject 1s singular in (1a) and plural in (1b) is indicated only by the numeral (yi- and san- respectively). No other indication of number is given either on the verb (you) or in the subject noun itself (mao). Another fact worth noting is that the inventory of agreement features is generally small and constant across languages: person, number and sometimes gender in sub- ject- verb agreement, while DP intemal agreement may include, besides number and gender, case and definiteness. Similarly, the number of feature values is generally low!; number includes, for instance, a singular and a plural value, plus, in some languages such as Arabic, a dual value. Finally, from a diachronical point of view, we can observe that languages un- dergo a process of loss of inflection, and consequently of agreement, as attested, for instance, in the evolution of Old English into Modern English, In Old English, nouns and adjectives are inflected for four cases and two num- bers; there are also three genders, while verbs have different endings for persons, numbers, tenses and moods, The inflectional paradigm of an Old English noun and verb are illustrated in (2a) and (2b) respectively: (2) a, Stan (stone m.) Old English SING PLUR N stn stiines G stnes stina D stine stnum A stan stanas. 1 A clear exception to this generalization is attested in Bantu languages which show an ar- ticulated system of noun classes. This will be illustrated and discussed in Section 5Constraining checking: on @features representation and their parametric variation . Drifan (drive) je drife we diifad dudrifest’ ge drifad he diifes hie drifad Of this rich inflectional system, only a number distinction survives in noun in- flection in Modern English, whose lexical verbs show a bi-partite indication of tense and an indication of third person singular only in the present tense While the fact that the inventory of features is almost constant across languages argues in favour of a universal set of features, eventually corresponding to a univer- sal set and/or hierarchy of functional projections,2 the facts that languages have a very restricted set of features and feature values and that this set is diachronically reduced seem to suggest that inflection and/or agreement are costly processes. ‘One obvious reason why features are costly is that they need to be checked, and checking entails application of Move, which is a costly operation for the grammar, In what follows, we'll expand this line of reasoning arguing for other reasons that explain why agreement is costly for the grammar and which feature array is less costly, given a model of agreement like the one proposed by Chomsky (1995b). Variation across languages will be examined and discussed in this perspective. Even if we restrict our attention to subject-verb agreement, in fact, we notice a cer tain range of cross-linguistic variation which concerns the number of features and feature values. Furthermore, variation can occur within the same language according to the constituent involved and the relative order of constituents, In semitic languages, for instance, the verb agrees with the subject in number, gender and person, as shown in (3) taken from Standard Arabic: 3) a Lrajul-u daxal-a Standard Arabic DETmanMNomS enterPF-3MS the man entered 2 1 am basically extending to agreement features an observation made by Cinque (1997: 219; fn. 11) in favour of a universal hierarchy of aspectual projections. Cingue (op. cit.) observes that the same (restricted) functional notions appear to be expressible (either via head morphology or via adverbs) language after language, while other functional notions are not. “...] no language appears to have a mood expressing the fact that there is “bad weather’ (or “good weather’) though we can easily imagine a world where that would be possible.” [Cingue, 1997: 219 fn. 11]. assume that the same argument holds for agreement features. 3 This is true when the verb follows the subject. With postverbal subjects we can have lack of number agreement on the verb. See Section 3 (ex. (13)) and Section 5 (ex. (33) 4BElisa Di Domenico b. L-mar’-at-u_ daxal-at DETwomanFNom$ _cnterPF-3 the woman entered ¢, I-mar’-at-aani daxal-ataa DETwomanFNom? enter?F-3FP the women entered In Romance (here exemplified by alan) and Turkish, instead, only number and person are subject/verb agreement features. as shown in (4) and (5) respectively: (4) a. Puomo entro Italian DETMS manMS_ enterPF-38 the man entered b. ladonna entrd DET.FS womanFS enterPF-35 the woman entered c. le donne entrarono DET FP womanFP enterPF-3P (8) a. Ab gehyor Turkish ‘Ali comePCont All is coming b. Selma geliyor Selma comePCont Selma is coming c. Ali ve Selma geliyorlar Ali and Selma comePContPl Ali and Selma are coming But while in Turkish there is no gender at all, Italian nouns, adjectives and 3rd person pronouns are inflected for gender: Italian, thus, does not employ one of its features in subject-verb agreement. A model of the optimal features representation will have to leave some space for the variation observed across languages. In Section 2 we'll briefly present the minimalist model of agreement trying to point out its advantages with respect to other approaches, but also the problems it poses. In Section 3 we'll present a general proposal concerning features representa- tion in order to avoid these problems, and some evidence. In Section 4 we'll discu the evidence and sharpen the proposal particularly with respect to the nature of fea- 44Consiraining checking: on ®-features representation and their parametric variation tures. In Section 5 we'll deal with some residual questions and alternative solutions attested in natural languages. 2 Minimalism and agreement Inflection and agreement have been accounted for in various Ways. Theories of in- fection have varied in considering it a lexical or a syntactic process (see Spencer (1991) for a review), while theories of agreement can be roughly divided into ‘direc- onal’ and ‘non-ditectional” ones. Directional approaches to agreement (see for in- stance Lehman (1982) and Nichols (1985)) maintain, more or less explicitly, that there tsa controller item and a target one. The controller endows the target With its features. Non-directional approaches, such as Chomsky (19956), maintain that features are assigned independently on the different lexical items, which subsequently agree. A third approach is the one proposed in the HPSG framework (see Pollard and Sag (1994). In this model, agreement is obtained through a process of ‘unification’ or feature ‘inheritance’. In a mode] of grammar which distinguishes independent components such as lexicon, computational component and interpretive component, agreement is to be considered a syntactic process, i.e. a process that takes place in the computational component, That being said, for such a model of grammar directional approaches pose, according to me, at least two main problemas First, the notion of ‘controller’ is not syntactically grounded since it does not nee- essarily correspond to the notion of head, as in the case of subject-verb agreement.4 Second, directional approaches presuppose a syntactic endowment of features, which is contrary to Emonds’ (1976) hypothesis that syntax is structure preserving. This latter problem concems unification approaches as well, at least as far as the mechanism of feature inheritance is concerned, Non-directional approaches present an advantage in this respect, but the arbitrary selection of features they entail poses additional problems. If features are selected arbitrarily, they need to be checked. ‘As we have already noticed in Section 1, feature checking forces movement, and movement is a costly operation. But the arbitrary selection of features in the numeration poses a further, serious problem in that it generates an ‘exponential blow-up’ in the evaluation of deriva- tions, certainly an undesirable consequence. 4 Gazdar et al. (tS) give a semantically grounded definition of controller and target, where the controller 1s an argument and the target is a functor. A semantically based defi- nition, however, does not solve the problem. 45Elisa Di Domenico For instance, given a language L with three features (e.g. Person, Number, Gen- der) and three values for each feature (e.g. 1,23; S(ingular), D(wal), P(lural); M(asculine), F(eminine), N(euter)), and considering solely verb features checking, only one derivation out of 27 will converge since there are indeed 27 possible fea- ture sets which can be chosen for the verb, as illustrated in (6): (6) Language I. Person: 1,2,3 Number: S(ingular), D(ual), P(tural) Gender: M(asculine), F(eminine), N(euter) {LSM} {LSE} {1,S.N} {LPM} {LPF} {LPN} {LM} {1,D.F} {1,D.N} (2,S,M} {2,S,F} {2,S.N} {2,D,M} (2,D,F} (2,D.N} {2,2.M} {2,P,F} {2,P,.N} GSM} BSF} BSN} GDM) BDF} 3.D,N} Only one of these 27 sets will match with the feature set of AGR: this means that, in the worst case, 26 derivations will crash before the convergent one is met. The same 27 possibilities will hold for the subject, and so, in the end, the convergent derivation will be one out of 54, A similar problem held implicitly in previous accounts of agreement such as Chomsky (1981; 1986). Within this proposal, the subject shares its features with INFL when it is in its Spec position, while the verb acquires the features of INFL adjoining to it.5 It is true that in this approach the computational cost is lower than in the minimalist approach: the verb, in fact, acquires its feature specification from INEL, and thus the problem of non- matching feature sets is limited to the pair sub- jecvINFL (27 possibilities instead of 27+27).6 The reasons why this approach changed to the minimalist one are various and it’s beyond the scope of this paper to discuss this issue, I'll just mention the problem that for languages like English, where there is no evidence of overt verb movement, it was necessary to postulate « rightward movement of the feature set of INFL to the verb (affix hopping), which was contrary to a generalized leftward movement. Furthermore, I think that it 1s advantage for a grammar to have a unique agreement procedure (as in the minimal- ist approach) instead of two different procedures, one for the subject and one for the verb, as in the Government and Binding framework In the minimalist approach both the verb and the subject come out of the lexicon with a full specification of fea- 5. Subsequently, after work of Pollock (1959) INFL is splitted into T and AGR (see Belletti (forthcoming) for a review). 6 Adopting a sugyestion of Luigi Rizzi (pe.), I consider the ‘sharing of features’ of Chom- sky (1981) as a sort of checking. 46Constraining checking: on @-features representation and their parametric variation tures which are then checked syntactically in the appropriate configuration with the features of the functional head. It is the functional head that drives the operation, which is characterized precisely as attraction by the functional head of the features of the lexical head (‘Attract F?), ‘Nevertheless, the problem of the computational complexity in the evaluation of derivations in the minimalist approach to agreement remains. It is true, however, that within this approach the already discussed observation (see Section 1) that ®- features and ®- feature values in natural languages are few, and that languages undergo a process of loss of inflection (which we have taken as suggesting that features are costly for the grammar) receives a natural explanation. It is clear that each feature, and each feature value increases exponentially the compu- tational complexity of the checking procedure. Thus, a language L” with three fea- tures as language L, but with two values for each feature instead of three, will have only 8 possible feature sels, instead of 27: (7) Language L’ Person: 1,2 Number: S, P Gender: F,M {SM} {LSF) {1PM} (1,P,F) {2M} {2,S,F) {2,.M) {LPF} But maybe it is also possible to constrain the checking procedure itself to avoid this exponential blow-up, as advocated by Chomsky himself: (8) “We want to formulate economy conditions that avoid ‘exponential blow-up’ in construction and evaluation of derivations” [Chomsky, 199Sb: 228]. ‘This is what we'll try to do in Section 3. 3. Constraining checking In order to identify an architecture for the checking procedure which is optimal with respect to the problem outlined in Section 2, we'll deal with some basic ques- tions, Let's call ®-P the syntactic projection of - features, and let's imagine it at- tracts @- features as a magnet attracts iron; now the questions are: - What is the optimal structure of features in lexical heads and in ®-P? - What are the primitives on which Attract F operates? As far as the first question is concerned, there is not much to say about feature structure in lexical heads: they must obligatorily occur all together in the lexical 47Elisa Di Domenico item. But what about ©-P? Is it a single projection in which all ©-features are pro- jected together, or is it split into different nodes? Stemming from the Split-Infl hypothesis (Pollock 1989), proposals have varied along the years with respect to the structure of AGR. More recently, Chomsky (1995b) argues in favour of the climination of AGR projections, as individual posi- tions, from clause structure. Their @-features are in fact not interpretable at LF. They are consequently assumed to be part of other projections: AGR-S, namely, is considered part of T. The second question can be paraphrased as follows: Does Attract F operates on categorial features such as Number, Gender, Person, or rather directly on their val- ues, i.e, Singular, Plural Feminine etc? The question is not trivial and addresses an- other interesting aspect of recent theory of phrase structure ( homsky (199Sa) ar- gues in favour of the elimination of categorial labels, assuming that it is the head itself that determines the label. Let’s take as null hypothesis the idea that Attract F operates on categorial labels and that ©-P is a unique projection. In our language L, all the 27 combinations in (6) will be attractable since all have Person, Number and Gender. But if Attract F operates directly on feature values, the range of attractable combi- nations will be reduced. Namely, instead of 27, the attractable combinations will be 19. If ©-P has, for instance, the values {1,S,M}, it will attract only the combinations in (9): (9) @P= {1,S,M} {SM} {SF} {1,S.N} {1PM} {LPF} {PN} {3,P,.M} {IDM} {LDF} {1D.N} (2.S.M} {2,S.F} {2.8} {2,D.M} {3,S.M} {3,S,F} {3,S.N} {2PM} {3,D.M} There will be in fact 8 excluded combinations: {2.D,F},{2.D,N}.{2,P,F}, 12.P.N}, {3,D.F}, {3,D.N}, (3,P,F},{3,P,N}. None of their values is compatible with the values in ©-P and so the whole set won't be considered. Besides other reasons, a bare phrase structure is thus advantageous also for the checking procedure, in that it initially excludes some combinations of features, namely 8 out of 27 in our example. This also means that objects like Person, Num- ber or Gender are not in the grammar. The procedure, however, will be even more considerably simplified if features are not grouped into a single projection, but are projected one per head. If features are projected one per head, considering our language L again, the at- tractable combinations will only be 9. Let’s imagine that -P has again the values in 4BConstraining checkin on @-features representation and their parametric variation (9), ie. {1,S,M}, and that we have M at the initial stage as shown in (10).7 The at- tractable combinations will be {1,D,M}, {1,P,M}, {2,SM}, (2,D.M}, {2,P,M}, (3,S.M}, (1,S.M}, (3,D,M}, (3,P,M}. With S in the second stage the set will be reduced to {1,S,M}, {2,S,M}, {3,S,M}, with the subsequent idenuification of {1,,M} with 1] in the final stage (10) ®P= [1[S[Mq1] [M] {1,D.M}, (1,P,.M}, (2,8,M}, (2,D.M}, {2,P,.M}, (3.5, (3,P,M}. [s} {1,S,M}, (2,S.M}, 3,8,M} a {1,8,M} The projection of one feature per head seems thus the optimal solution, in that it reduces considerably the complexity of the checking procedure. Instead of checking (a maximum of) 27 combinations, only 9 of them will be checked initially, and in a second stage 3 of this initial set will be checked again, for a total of (maximum) 12 checking operations. It is perhaps worth noting that in a language where all features have the same number of values, as in our sample language L, the order in which features are pro~ jected does not influence the procedure: the initial range will be of 9 derivations whether we have M or i or S in the initial stage. : But in another language L’’, in which we have for instance 4 values for person and 2 for number, it will be better, from our point of view, to have person at the bot- tom, since it is the feature with more values.8 Namely, it will cut into four instead of two subsets the set of possible combinations, and only 1/4 of the possible combina- tions will be initially attracted. If this is true, then it follows that some variation across languages should be ex- pected with respect to the order of agreement-related projections. This might be in line with Cinque (1997) who observes that the limited cases of apparent variation in the order of functional projections seem to involve agreement and negation in rela~ tion to other functional heads. }, (1,8.M}, {3,D,.M}, 7 Considering the construction of derivations a bottom-up procedure, I take the “initial stage’ to be the lower projection. 8 I till use labels such as ‘number’ or ‘person’ but only as convenient descriptive tools. 49Elisa Di Domenico We'lt come back to this topic in the next sections. Let’s conclude this section with the observation that the finding that the projection of one feature per head re- duces exponentially the complexity of the checking procedure is in contrast with assumptions such as Chomsky (1995b) who are in favour of ‘a single projections for Tense and AGR-S. This finding is rather in line with a series of works which assume a more articu: lated structure of AGR related projections, Many works, starting from Rizzi (1986), propose the existence of multiple subject positions. More recently, Cinque (1997) for instance, assumes that there are several DP-related projections even in the higher portion of clause structure. He observes that subjects in Italian must be higher than low adverbs like mica, but can be found to the left or to the right of higher adverbs like rapidamente, as shown in (11) (11) a, Maria mica prende il treno Maria not takes the train a’.*Mica Maria prende il reno ‘Not Maria takes the train b. Rapidamente Gianni alzi di nuovo il braccio Quickly Gianni raised again his arm b’ Gianni rapidamente alz6 di nuovo il braccio. [Cingue, 1997: 183-185] Gianni quickly raised again his arm Cardinaletti (1996), on the same line, assumes that there are two preverbal sub- ject positions, mainly on the basis of the distribution of weak subject pronouns. These pronouns cannot be left dislocated, but, in Italian, they can be separated from the verb by adverbs and parentheticals. Since parentheticals cannot adjoin to X’ but only to XP, the pronoun cannot occur in one and the same position with the verb. At the same time, it cannot be left-distocated.9 The problem can be solved under the assumption (universally valid) that there is more than one preverbal subject position. The two subject positions are specialized for different kinds of subjects. Similarly, Shlonsky (1997) assumes multiple subject positions analyzing the position of differ- ent kinds of subjects with respect to negation in copular constructions and benoni sentences in Modem Hebrew. Other works not only postulate multiple subject-related positions, but pursue the idea that these positions are headed by different features. Shlonsky (1989), for instance, proposes a structure like the one in (12): (12) {Person P[TenseP[NumberP[GenderP]]] 9 Similar observations hold for subjects in English and French. 50Constraining checking: on @.features representation and their parametric variation In Standard Arabic with a non-pronominal postverbal subject (which is of neces- sity third person) the verb is not inflected for number, but only for gender, as shown in (13a). There is also a person inflection, which corresponds however, according to the author, to a non-person, given the impersonal value that third person has in Ara- bic and other semitic languages. With a preverbal subject the verb is inflected also for number, as in (13b): (13) a ‘akalea Le ‘awlaad —1-TaCaam Standard Arabic eatPassM Det-boy-MPI Det-food ‘The boys ate the food b. Quitu ‘inna I ‘awlaad ‘akal-w — 1-TaCaam, sayl$ that Det-boy-MPI eatPassMPI Det-food I said that the boys ate the food {Shlonsky, 1989, 3] This shows, according to the author, that you can have gender without number and person, Furthermore, Modern Hebrew has a verbal form called benoni which corre- sponds to a present tense and is inflected only for gender and number, but not for person, as shown in (14a). The past tense is inflected for person, number and gen- der, as in (14b): (14) a, ata Somer al ha-xacilim Modern Hebrew you guard-MS on Det-cggplants You are guarding the eggplants b. ata Samar-ta al ha-xacilim you guard-Pass-2MS on Det-eggplants You were guarding the eggplants [ibidem p. 5] This shows that gender and number can be present without person. The data in (13) and (14) taken together suggest an implicational hierarchy of features like (15), from which the clause structure in (12) is derived: (15) Implicational Hierarchy of Agreement Features -Ifa verb is inflected for number then it is also inflected for gender -Ifa verb is inflected for person then itis also inflected for number [ibidem: 6] I see two main problems in this analysis. First, examples like (13a) cannot be easily taken, according to me, as instances of absence of person, since there is in- deed an indication of person on the verb. Second, verbal forms like benoni are better characterized as participial forms which would then correspond to an AGR-O rather than to the lower part of AGR-S.Elisa Di Domenico Nevertheless, the basic idea is extremely interesting, and it has been developed in more recent works with different kinds of data. Poletto (to appear) and Manzini-Savoia (to appear) observe that different kinds of subject clitics in Northern Italian dialects occupy different positions, at least with respect to negation, as shown in (16) from Fiorentino: (16) @ un tu vieni Fiorentino not you come b. Ja un viene she not comes c. unvu—venithe not youP! come dle un vengano [Manzini-Savoia, to appear] they not come While the second person clitic (16a; 16c) follows negation, the third person clitic precedes it (16b; 16d). This leads the authors to postulate, independently, different subject positions. While Poletto proposes a differentiation in terms of a feature (#Hearer} [+ Participant}, Manzini-Savoia propose a differentiation of D/Person fea tures. The two proposals differ also in other aspects. Poletto maintains that AGR-S is, splitted into four distinct positions, two of which are higher than Negation and the position where the complementizer occurs, while Manzini-Savoia assume four dis- tinct positions interpolated with different Neg projections. It is beyond the scope of the present work to discuss these proposals and their differences in details. What is important to our concem is that both works go in the same direction, as the other works discussed above, suggesting that AGR-S is not a unique projection. Furthermore, these works suggest a differentiation of AGR pro Jections in terms of feature content. Finally, both Poletto and Manzini-Savoia propose a structure articulated along features that do not necessarily coincide with familiar notions such as gender, num- ber and person. We'll develop this topic in Section 4. To sum up, the works reported here suggest that there is indeed an articulated structure of ®-P, In this section, we have shown why this structure is more conven- ient than a single projection: The projection of one bare feature per head reduces 52Constraining checkin on @-features representation and their parametric variation exponentially the complexity of the checking procedure, which would otherwise lead to an exponential blow-up in the evaluation of derivations.10 4. On the nature of the basic features of D-P ‘As we have scen, it is essential not only that features are projected one per head, but also that these features be bare: this means that they will not be Person or Num- ber, but Singular, Plural, 1, 2 ete. If itis so, then we'll have an alternative projection of features along the same node. 1,2,3 for instance, will be projected alternatively under one projection, Singular, Plural, Dual under another etc. Such a hypothesis, however, is incompatible with data like (16), where we see that subject clities of 3rd and 2nd person occupy different positions. Furthermore, if values are projected alternatively, we'll have different types of functional projections for different derivations These facts together suggest that, although the primitives on which Attract F op- erates are not categorial features for the reasons outlined in Section 3, they can nev- ertheless be feature values We have then to imagine some other kind of basic features which are neither categories nor values, In order to do this, I will consider two independent intuitions. The first one is the already mentioned proposal of Poletto (to appear) which characterizes subject clitics in Northern Italian Dialects as «itterentiated in terms of features like [+Hearer] [ Participant]. The other proposal I'll consider is in Cinque (1997), who argues in favour of a universal array of functional projections present in every sentence, with cach head coming with a marked and a default value. Intuitively, it is clear that the values traditionally defining a feature can be grouped into partially inclusive sub-sets: Ist and 2nd person, for instance, are both [+ Participant] as opposed to 3 person. I propose that (functional) projections are headed by partially inclusive subsets of features which can be switched positively or negatively. As far as the notion of markedness is concerned, I assume that, since the system is built to minimize check- ing, a value is marked if it attracis the widest range of derivation. The features I propose to characterize traditional labels such as Person, Number and Gender are given in (17), (18) and (19) respectively:!1 10 {tis worth noting that while I agree with Nash-Rouveret (1996) in splitting AGR, my pro- posal is radically difterent from theirs because I crucially assume that projections are not contentless, i. not proxies. 11 But see Section § for a partial reconsideration of gender. 53Elisa Di Domenico (17) Person : [+ Participant] [+ Speaker] (18) Number: [+ Plural] [+ Dual] (19) Gender : [+ Feminine] The features in each category will be ordered in such a way that the one at the bottom will attract the narrowest range of derivations. This means that, for instance, [Speaker] will be lower than [Participant], and [Dual] will be lower than [Plural].!2 A certain degree of variation is to be assumed as far as the kind and number of basic features is concerned: there is probably no {+ Dual] in Romance. And even restricting our attention to a single language, we can imagine that a node is projected only if necessary: [Participant]. for instance, is projected only in case of [Speaker] negatively switched. But if we want to pursue the idea ofa univer- sal hierarchy of functional projections we can imagine that an unnecessary node ex- ists but is not activated, i.e. contains no + value. The data in (16) seem to suggest that this second option is valid, since 3rd per- son is higher than 2nd person. Assuming the second solution, the characterization of Ist, 2nd and 3rd person in terms of features like (+ Participant] [+ Speaker] will be the one in (20) (20) a. Ist Person b. 2nd Person ©. 3rd Person {[Participant[+Speaker]] [+Participant (:Speaker]] _ {-Participant{Speaker]} In such a structure, a third person subject clitic, for instance, will be merged un- der [-Participant], since [Speaker] is not activated. A second person subject clitic, conversely, will be merged under {-Speaker]. In varieties like Fiorentino, since its position is lower than the position of the 3rd person pronoun (see (16)), we have to assume that it raises covertly to [+Participant]. The characterization of features in (20) points to an asymmetry between Ist and 3rd person one side, and 2nd person on the other. 2nd person is the marked form because it initially attracts the widest range of derivations: in our language L, [-Speaker] will attract 18 derivations, while both [-Participant] and [+Speaker] will attract 9 derivations. If this characterization is correct, and 2nd person is the marked form, than we find an interesting suggestion in correlation to Jakobson’s generaliza- tion: 12 See below, at the end of this section, for the problem of the relative order of features like {Participant] (or (Speaker]) with respect to [Plural} (or [Dual]}. 34Constraining checking: on D-features representation and their parametric variation (21) Jakobson’s generalization Zero morphology typically occurs with the unmarked members of the catego- ries [Jakobson 1971/1939 reported in Cinque 1997: 214] As reported in (21), Jakobson observes that the marked members of categories are usually morphologically signalled. Now, as noted by Renzi-Vanelli (1983), the most widespread subject clitic in Northern Italian Dialect is the 2nd person (singu- lar)elitic: (22) Renzi-Vanelli (1983) generalization Ifa variety has only one subject clitic, it is a second person singular!3 In fine with Renzi-Vanelli’s generalization, Rizzi (1993) observes that in Italian, when the inflectional paradigm is not rich, as in the subjunctive mood, a null subject is impossible in the second person (23b) while it is tolerated in the first person (23a) and perfectly acceptable in the third person (23c): (23) a, Pensano che ?(i0) venga domani They think (I)’d come tomorrow ’b, Pensano che * (tu) venga domani They think (you)'d come tomorrow c. Pensano che (Iuislei) venga domani ‘They think (he/ she) would come tomorrow [Rizzi, 1993] Rizzi’s observation and Renzi-Vanelli (1983) generalization point out that sec- ond person is the most frequently specified member of the category of person. If Ja- kobson is right in maintaining that the most specified member is the marked one, then we have indirect evidence that our characterization of second person is correct: as shown in (20), second person is the most marked member, because it attracts i tially the widest range of derivations. To sum up, the features I have proposed in (17), (18) and (19) solve the prob- Jems that both categorial features and bare feature values posed to a system meant to optimize checking. They also fit with data like (16) that strongly suggest that there is more than one projection for person. Finally, they are able to capture and explain not only the difference between Ist/2nd person and 3rd person, but also the asym- metry between Ist/3rd person and 2nd person observed by Renzi-Vanelli (1983) and Rizzi (1993). 13 Maybe the generalization as formulated in (22) is too strong. Savoia-Manzim (to appear) report that in the variety of Vermiglio there is only a 3rd person cline Renzi-Vanelli’s generalization can be however maintained in a weaker form, i.c. meanimg simply that the 2nd person clitic is the most widespread, 55Elisa Di Domenico There are however many questions unsolved. The first one concerns the relative order of projections like [Participant] (or {Speaker]) with respect to [Plural] (or (Dual]). While the system induces an ordering of [Speaker] with respect to [Participant] (and of [Dual] with respect to [Plural]), the order of [Participant] with respect to [Plural] seems to be unimportant to constrain checking, in a system where features have binary values as the one we have developed in this section.|4 One might argue that order is language specific. Poletto (op. cit.), for instance. observes that subject clitics of the same morpho- logical type can be found in distinct positions in different dialects. This observation can bé interpreted as due to the different relative ordering of projections. ‘The other possibility is to explain this variation in other ways,!5 and maintain that the order of projections is universal, certainly a more desirable solution from a theoretical point of view. In any case, what is to be said is that the system here developed does not induce ordering of all projections, but only of some of them. ‘Another question concems the status of certain features which show a wide cross-linguistic variation, such as gender. We'll examine this and related questions in Section 5. 5. Parasitic features and other ways to constrain checking We have seen in Section I that the presence! absence of gender is subject to cross-linguistic variation, Some languages with agrcement don’t have gender, like Turkish, Other languages with gender don’t employ it for subject-ver agreement, like Romance languages. Semitic languages, conversely, have gender also in sub- ject-verb agreement. Another interesting fact not mentioned in Section 1 is that while in Romance gender is limited to 3rd person, in Semitic we have gender in the 2nd person as well as in the 3rd, as shown in (24): 14 In Section 3 we observed that if features have different values, the feature at the bottom must be the one with more values, because it reduces more considerably the range of at- tractable derivations. 15. See Manzini-Savoia (op. cit.) for a proposal in this direction with the use of different Neg projections, 56Constraining checking: on features representation and their parametric variation (24) Verb jls (to seat) Singular Dual Plural Masc Fem = Mase Fem Mase Fem Ip jalas-tu jalas-na 2p jalas-ta_jalas-ti jalas-tumaa jalas-tum jalas-tunna 3p jalas-ajalas-at_ jalas-an jalas-ataa jalas-uu jalas-na ‘These facts suggest the idea that gender is projected in a different way according to the language considered. We can mamtain, for instance, that in Semitic gender is selected by [+Participant], thus explaining why it is a feature of 2nd and 3rd person, but not of Ist person. The structure for Ist person singular and for 2nd person singular femi- nine is given in (25): 25) a. 1S b. 2,F.S [-PluralfParticipant [+Speaker[]] [- Plural [+Feminine [+Participant[-Speaker]]]} In Romance, as we have seen, there is no gender in subject-verb agreement. I ar- gue that this is due to the fact that there is no [+ Feminine] projection in the clause structure of Romance, or, if there is such a projection, it is not activated. Since in Romance there is a DP-internal gender agreement, we might postulate a [# Feminine] projection in the Romance DP, and hence an asymmetry between DP and CP structure. Gender is in 3rd person only, and so we have to think that it is se~ lected by (-Participant). The structure of the Romance -P will be like (26a) for CP and like (26b) for DP in case of 3rd person singular (feminine): (26) a. 3,8 b. 3SF [-Plural[-Participant[Speaker]]][-Plural[-+Feminine (-Participant{Speaker]]]] There are, however, some arguments that suggest that gender is not a head. One argument to think that gender is not a head can be taken ftom Cinque (1995). Cinque assumes that the position of adjectives in Germanic and Romance is the same and that the different superficial order, that can be observed in (27), is due to noun movement to check ®-features, which are strong in Romance and weak in Germanic: (27) a. L’invasione italiana dell’ Albania / *L’italiana invasione dell’ Albania b. The italian invasion of Albania / *The invasion italian of Albania [Cinque, 1995] Cinque observes, however, that this movement is only one head higher, and not two as it should be to check mumber and gender.Elisa Di Domenico The data receive a natural explanation if gender and number are projected under the same head. Another evidence that supports this idea comes from production er- rors. Garcia - Albea - del Viso - Igoa (1989) collected a corpus of slips of the tongue in Spanish. They found that errors like (28a) are much more frequent than errors like (28b): (28) a. Un duro de veinte monedas a’.Una moneda de veinte duros A coin worth twenty ‘duros’ b. Hay médica de huelgos b’.Hay huelga de médicos ‘There’s a doctor’s strike (Garcia - Albea - del Viso - Igoa, 1989} In (28a) is shown a movement error with number stranding: while the stem is in the wrong place, the indication of number is left in its original place. In (28b) there is a movement error with gender and number stranding. The authors never found errors with only gender stranding: this, again, suggests that gender is not a head. If gender is not a head, one can maintain that 11s a parasitic feature. ‘The difference, in our system is that a head will add a further step to the check- ing procedure, while a parasitic feature will mnerease the number of feature values, entailing that the projection cannot be set in terms of binary values . In a language like Italian, for instance, where gender and number have two val- uucs cach, if they are checked together under the same projection they will give rise to four possible combinations.16 The structure of DP and of CP is thus similar in Romance: the only difference is that DP allows a parasitic feature, i.e. gender. The difference between Romance languages and Turkish (where there is no gen- der at all) can be stated in these terms: While Romance languages allow parasitic features, languages like Turkish do not Interestingly, a whole group of languages behaves like Turkish, namely all lan- guages of the agglutinating type, as stated in Renault’s generalization in (29): (29) Si une langue est du type agglutnant, alors elle ne peut pas avoir de genre grammatical (Renault, 1987: 113] Renault’s generalization can be then explained as in (30): (30) Agglutinating languages do not allow parasitic features. 16 As I have discussed in details in Di Domenico (1997) there are reasons to believe that gender is indeed a parasitic feature of Num or of N according to the kind of gender (vari- able or fixed) of the noun. This js also suggested by the data in (28). 58Constraining checking: on P-features representation and their parametric variation Few words need to be said on parasitic features. On one side they are a way to constrain checking, in that they do not add a further step in the derivation. But on the other side they widen the range of the attractable derivations, for the reasons outlined in Section 3. In some cases, however, they are the most economic strategy, i.e, when feature values are many, as in the case of Bantu languages, where there is arich system of noun classes: (31) Noun Classes in Swahili [Polomé, 1984] Prefix ‘Nouns included m___| humans wa [plural of class 1 mr long or thin objects, trees mi plural of class 3 ji- paired or grouped objects, fruits ma-___[ plural or collective of class 5 ki- instruments. vie plural of class 7 9 (ng _| miscellaneous, animals plural of class 9 parts of the body (take cl.10 as plural) in proto -bantu,diminutives plural of class 12 ue abstracts ku-____| infinitives definite location indefinite location location inside a place In this case we have to imagine a single projection which includes number and gender whose value is not stated in binary terms such as + or -. To minimize check- ing, this projection should be the lowest, and, since Bantu languages have gender in subject-verb agreement, should be present both in the DP and in the clause. At an opposite side is the organization that we find in Somali. As shown in (32), when a noun changes its number, it must also change its gender: 59Elisa Di Domenico (32) a. dibi dibiye Somali bullMs bullsFP! b. layr layro airFS airsMPL c. aabbé abbayadl fatherMS, fathersEPI 4. bug buugig bookMS booksFPL (Puglielli-Siyad, 1984) ‘The phenomenon, known as ‘polarity’ (Meinhof, 1912), shows, according to me, that in Somali gender and number are clearly in the same projection, but the projec- tion maintains its binary system of values: if you change number, you are forced to change gender. The phenomenon is reflected in DP-internal (noun-determiner) as well as in subject-verb agreement: hence we have to hypothesize that this projection is both in the DP and in the CP. One might wonder whether the characterization of Semitic given in (26) is cor- rect of if it would be necessary to reconsider it in the terms used for Romance, Bantu and Somali, i. with gender as a parasitic feature. The problem is that Semitic seem to be the only languages with a projection specialized for gender, at least as tar as the CP is concerned. Some evidence, which I unfortunately cannot provide at the moment, would be required to establish whether Semitic languages have this extra projection or not, What can be said at this moment is that, if gender is parasitic also in Semitic, it can- not be parasitic on number, at least in the CP. This not only because it appears also in the 2nd person, and thus it seems connected to [+Participant), but also because gender is attested in cases of absence of number agreernent, i.e. the already men- tioned partial agreement with post- verbal subject (33) a. bmar’-at-u daxal-at ¢. daxaleat I-mar’-at-u DETwomanFNom$ enterPF-3FS _enterPF-3F§ DETwomanFNomS the woman entered entered the woman b. Imar’-at-aani daxal-ataa d,_daxal-at |-mar’-at-aani DETwomunFNomP enterPF-3FS —_enterPF-3FS DETwomanFNomP the women entered entered the women 6. Conclusions In this paper we have discussed some advantages and disadvantages of the minimalist approach to agreement, pointing out that the arbitrary selection of fea- tures, with the subsequent necessity of checking, complicates considerably the 60Constraining checking: on @-features representation and their parametric variation evaluational procedure. This complication, as we have shown, is exponentially re- lated to the number of agreement features and of their values. While this might be the reason why languages generally constrain the number of features and feature values, we have proposed a clausal architecture that can reduce exponentially the complexity of the checking procedure. The system is built on two basic assumptions: i) Attract F does not operate on categorial features ii) Features are projected one per head We have proposed a set of basic, partially inclusive, features which do not coincide either with categorial features such as Number and Person or with their values (1,2 etc.); these features can be set in binary terms with a + or a— value, or contain no value and hence be inactivated as attractors. Features are ordered in such a Way that the fea- ture at the bottom is the one that attracts the narrowest range of derivations, Finally we have examined some evidence and some cross-linguistic variation, arguing that natural languages can also offer partially alternative solutions. Many problems remain, and further evidence needs to be gathered: we leave these tasks to further research. References Belletti, A. (forthcoming), Agreement Projections, to appear in M. Baltin and C. Collins (eds), The Handbook of Syntactic Theory, Blackwell Cardinaleti, A. (1996), Subjects and Clause Structure, University of Venice Working Papers in Linguistics,611 Chomsky, N. (1981), Lectures on Government and Bindvre, Dordrecht, Foris. Chomsky, N. (1986a), Barriers, Cambridge, Mass., The MI! Press. ‘Chomsky, N. (1991), Some notes on the economy of derivation and representation, in G. Freidin (ed.), Principles and Parameters in Universal Grammar, Cambridge, Mass., The MIT Press. Chomsky, N. (1995a), Bare Phrase Structure, in G. Webelhuth (ed.), Government and Binding and the Minimalist Program, Blackwell, Oxford. Chomsky. N. (1995b), The minimalist program, Cambridge, Mass., The MIT Press. Cinque, G (1995), On the evidence for partial N movement in the Romance DP, in G. Cinque, italian Syntax and Universal Grammar, Cambridge, Cambridge University Press Cinque, G. (1997), Adverbs and functional heads. A cross-linguistic perspective. University of Venice Working Papers in Linguistics, 7,1-2 Di Domenico, E, (1997), Per una teoria del genere grammaticale, Padova, Unipress. Emonds, J. (1976), 4 transformational Approach to English Syntax, Academic Press. Ferguson, Ch.A. - Barlow, M. (1988), ‘Introduction’, in M. Barlow and Ch.A. Ferguson (eds.) Agreement in Natural Languages, Stanford: CSLI. 61Elisa Di Domenico Garcia-Albea, J.E. - del Viso, S. - Igoa, J. (1989), ‘Movement Errors and Levels of Processing in Sentence Production’, Journal of Psycholinguistic Research, 18/1: 145-161 Gazdar, G. - Klein, E. - Pullum, G. - Sag, 1. (1985), Generalizes Phrase Structure Grammar, OX- ford, Blackwell Haiman, J. (1985), Natural Syntax, Cambrudae Cambr «leo University Press. Jakobson, R, (1971), [originally 1939}, Signe zero, in Selected Writings, vol. II. Mouton, The Hogue: 211-219. Jespersen, O. (1922), Language: lts Nature, Development and Origin, New York, Macmillan Lehman, C. (1982), ‘Universal and typological aspects of agreement’. In H. Seiler and F.J. Sta- chowiak (eds.), Apprehension: das sprachliche Erfassen von Gegenstinden, Teil Il, Tie bingen, Narr. Mancini, MR. - Savoia, L. (to appear), Parameters of subject inflection in Northern Italian Dia- lects, Proceedings of the EPP Conference, Teomsa, June 1997. Meinhof, C. (1912), Die Sprachen der Hamicen, Hamburg, Abh, des Hamburgischen Kolonialin- stitus. Nash, L. - Rouveret, A. (1996), Proxy Categories in Phrase Structure Theory, Ms. Université Paris 8 Nichols, J. (1985), The directionality of agreement, Proceedings of the Eleventh Annual Meeting of the Berkeley Linguistic Society. Poletto, C. (to appear), The Intemal Structure of AgrS and Subject Clities, in H. van Riemséijk (ed.) Clluies in the Languages of Europe, Eurotyp Project Working Papers. Pollard, C. - Sag, C. (1994), Head Driven Phrase Structure Grammar, Chicago, University of Chicago Press. Pollock, J.-Y. (1989), “Verb Movement, UG and the Structure of IP", Linguistic Inquiry, 20: 365- 424. Polomé, E.C. (1967), Swahili Language Handbook. Washington DC, Centre for Applied Linguis- ties Puglielli, A. - Siyad, CM. (1984), ‘La flessione del nome’, Studi Somali, 5: 53-114, Renault, R. (1987), ‘Genre grammatical et typologie linguistique’, Bulletin de la societé de linguistique de Paris, 82: 69-117. Renzi, L. - Vanelli, L. (1983), I pronomi soggetto in aleune varieté romanze, Sorte’ linguistict in onore di Giovan Batista Pellegrini, Pisa, Pacini Rizzi, L. (1986), On the status of Subject Cities in Romance, In O Jueveli » C, Silva-Corvalan (€ds:), Studies in Romance Linguistics, Dordrecht, Fotis. Rizzi, L. (1993), Class Lectures, University of Geneva. Savoia, L. -Manzant, MLR. (to appear), [dialett italiani, Bologna, I! Mulino. Shlonsky, U. (1989), The Hierarchical Representation of Subject Verb Agreement, Ms. University of Hails Shlonsky, U. (1997), Les phrases copulatives et la position sujet, Talk delivered at the University ‘of Siena, December 1997. Spencer, A. (1991), Morphological theory: an introduction to word structure in generative gram- mar, Oxford, Blackwell. 62FEATURE MISMATCHES IN GREEK CLITIC LEFT-DISLOCATION CONSTRUCTIONS Angela RALLI— Manuel ESPANOL-ECHEVARRIA 0. Introduction A research strategy advocated by Boojj (1997) regarding the autonomy of mor- phology is to discover phenomena in which morphology is different from the other grammatical modules, ie., syntax and phonology. After a survey of Case mismatches between syntactic constituents that agree for other morphosyntactic features, e.g, Number and Gender, we propose that Case belongs to features which instantiate the autonomy of morphology and does not lend itself to simple reduction to syntax. Following the minimalist framework, we show that the overt manifestation of Case, the most structurally relevant of all inflectional features (as defined by Kury- lowicz, 1964), does not entirely depend on syntactic considerations (e.g., syntactically checked Case, cf. Chomsky (1995). In accordance with Ralli (1997, 1998), who claims that inflectional features are primarily morphological and, as such, are accounted for within morphology in a principled-way, we work on the assumption that Case marking is handled within a morphological module of grammar: the overt occurrence of a Case feature is closely related to morphological structure and very much constrained by lan- guage-dependent characteristics. However morphological Case affects with Syntax in several ways. It is shown that, in a language with a rich Case system like Greek, the syntactic manipulation of Case 13 very much affected by morphological constraints, although it must be as language mdependent as possible on the basis of the minimalist assumptions that syntax provides the bare essentials to the representation of the lan- guage faculty. This proposal brings addinonal support to Aronoff’s (1994) views (fur- ther elaborated by Ralli 1997, 1998 with respect to inflectional features) that morphol- ogy, although being independent, interacts with syntax on several aspects. Our claims are supported with data mostly taken from Greek, a language rich in Case marking. A typical example illustrating these claims is the phenomenon of Case mismatch in Clitic left-distocation (CLLD) constructions. 63Angela Rall J. The data 1.1. Case mismatches in clitic left-dislocation constructions Elaborating on Cingue (1990), Baker (1996) claims that dislocated constituents in languages, both polysynthetic and non-polysynthetic, form a non-movement cham with a unique pronominal in argument position. The conditions on this chain formation are stated as follows (cf. Baker 1996: 1 12) (1) The Chain Conditions X and Y may constitute a chain if: () Xc-commands Y. (ii) X and Y are coindexed (iii) There is no barrier containing Y but not X (iv) X and Y are nondistinct in morphosyntactic features (i.e., category, per- son, number, gender, case, etc.) According to the condition given in (liv), feature mismatches between left- dislocated constituents and the pronominals coindexed with them are not allowed. Baker, however, has observed some apparent feature mismatches between the two constituents in a polysynthetic language like Mohawk: (2) Sak wa-shukeni-k’ ne ra6-skare’ —-kanét-a-ku Sak FACT-MsS/1DO-see-PUNC ne MsP-friend _ town-@-in “Sak saw me with his girlfriend in town” (Lit.: “His girlfriend, Sak saw us two in town”) (From Baker 1996:122, where “FAC” stands for Factual, “PUNC" for Punctual, “s” for singular, “S” for Subject, “D” for dual, “M” for Masculine, “O” for Object, “1” for first person, and “P” for Possessor’). The occurrence of feature mismatches in polysynthetic languages is explained by Baker (1996) by the fact that, in these languages, the morphological expression of morphosyntactic features involved in the coindexation is not overt. Thus coindexa- tion becomes possible since there is no formal marking for these features. In a non-polysynthetic language like Greck, this explanation predicts that dislo- cated nominal constituents require a matching between theit morphosyntactic fea- tures and the features of pronominal elements coindexed with them since, in this language, nominals are overtly marked for Gender, Number and Case.’ Any feature Inflectional class is another feature that overtly marks the inflectional system of Greek nominals, but, as Ralli (1997, 1998) has shown, itis syntactically irrelevant.Feature mismatches in Greek clitic left-dislocation constructions mismatch between a dislocated DP and the pronominal element (clitic) in its chain would Icad to a violation of the condition (iv) in (1), ruling out the well-formedness of the CLLD construction. In Greek, however, there are counter examples to this prediction, particularly with respect to Case matching. Let us see some of these examples which are bor- rowed from works by Philippaki-Warburton & Stavrou (1986), Catsimali (1990), Philippaki- Warburton (1990), Tsimpli (1990), Alexiadou & Varlokosta (1996): (3) a. Opjos m’agapai, ton agapo whoever-NOM — me-ACC-loves, him-ACC love-1$ “| love whoever loves me” (or Whoever loves me, I love him) b. Opjon den grapsi —_sosta afto to thema, tha ton aporipso” whoever-ACC not write-PERF right this the topic FUT him-ACC discard- PERF “I will discard whoever does not write well this topic’ (4) 1 fitites, i kathijites tus agapane lus? the students-NOM, the professors-NOM them-ACC love-3PL all-ACC “The professors love all the students” (or The students, the professors love them ail) In (3a), the head of the dislocated free relative opjos occurs in nominative case, although it is coindexed with an accusative clitic, i., con. In (3b), the head of the dislocated free relative opjon has the same case (Accusative) with the coindexed cli- tic in the matrix clause, but it cannot be checked by the verb of the relative which As noted by Philippaki-Warburton & Stavrou (1986), in the matching cases of free rela- tives, the phrase raises to the head position , whereas in the non-matching ones, it remains in COMP. Cheila-Markopouiou (1991) has observed that the basic problem in this analy- sis is that it seems to involve improper movement from an AA’ to an A position, Generally, an attempt to relate the non-matching effect to the pro-drop parameter is not correct as pointed out by Alexiadou & Varlokosta too (1996). As noted by Tzartzanos, Case mismatches, such as the one illustrated below, are much better if some material (related to the dislocated constituent) intervenes between the dis- located constituent and the rest of the sentence: (OT kira-Rini tu Kritu, tu Duka j thigatera, xronia tis etimazun ta prikia (from Tzartzanos (1946: 264)) ‘The-NOM Irene-NOM of Kritos, the daughter of Dukas, years-ACC her-GEN pre- pare-3PL the dot “Mrs. Irene of Kritos, Duka’s daughter, they prepare her dot for years”. 65Aagela Ralli requires nominative case. In (4), the dislocated DP i fitites “the students” shows nominative case, but it is coindexed with an accusative clitic, i.e. tus “them.” These Case mismatches pose a problem for Baker’s (1996) unified account of dislocated constituents. in both polysynthetic and non-polysynthetic languages. Par- ticularly, any attempt to provide a syntactic explanation for the presence of nomina: tive marking for the dislocated DP in (4) is problematic since this nominative case is syntactically unjustified and remains unchecked contrary to recent minimalist claims that [-interpretable] features should be checked and crased before the C-I level. In fact, Alexiadou and Varlokosta (1996) have already pointed out that Case matching in constructions involving dislocated constituents does not concern abstract syntac- tic identity, but seems to be relevant to PF interface.* That Case marking and Case mismatch in Greek CLLD constructions cannot be accounted for within the strict limits of a syntactic analysis is also supported by the fact that in other syntactic contexts, ¢.g.. in contexts where the free relative or the DP occur in a non-dislocated position, the relative pronoun or the DP must bear the Case marker checked by the matrix verb, Compare the following examples: (5) a. *Agapo opjos m’agapai love-IS_ whoever-NOM_ me-ACC-loves b. Agapo opjon magapai love-1S whoever-ACC — me-ACC-loves “[love whoever loves me” (© a. #1 kathigites agapane oli i fitites the professors-NOM love all-NOM_the-NOM students-NOM b. 1 kathigites agapane olus tus fitites the professors-NOM love all-ACC the-ACC students-ACC “The professor love all the students” Moreover, constructions involving a leftward movement of the free relative or the DP, do not allow the Case mismatch observed in (3) and (4): That Free Relatives involving non-matching effects are left dislocated has also been claimed by Alexiadou & Varlokosta (1996). They argue that left-dislocated DPs are base generated and do not involve movement. They base their proposal, however, on Ma- rantz’s (1991) approach concerning Case Theory. They suppose that the case marker is added after Spell-Out and case realization depends more or less on syntactic considera- tions, that is on whose element’s complement doinain the NP appears at PF. Since the D of the dislocated DP (or the wh-element of the free relative) does not appear in the com- plement domain of any predicate, non-matching is possible. 66Feature mismatches in Greek clitic left-dislocation constructions (7) a. *OPIOS m’agapai, agapo whoever-NOM_ me-ACC-love-38. love-1S b. OPJON m’agapai, agapo whoever-ACC me-ACC-love-38love-IS (8) a *Oll 1 FITITES agapane I kathijites all-NOM the-NOM students Jove-3PL the-NOM professors-NOM “The professors love all the students” b.OLUS TUS FITITES —_agapane i kathijites all-ACC the-ACC students-ACC love-3PL the professors-NOM “The professors love all the students” (7) and (8) show that nominative DPs cannot appear in object position, and fronting due to Focus does not irigger the possibility of Case mismatch, as noted by Alexiadou & Varlokosta (1996), 2. The analysis In this paper, we claim that Case mismatches are mainly due to the fact that mor- phology, where word formation occurs, is independent from Syntax. We follow the approach according to which words are firstly built within Vorphology and enter Syntax already marked for their morphological features, Case being one of these features (cf., among others, Chomsky 1995). In our analysis, we take advantage of the following assumptions put forward by Ralli (1997): a) ina modularly built grammar (cf. Di Sciullo 1996), morphology is an independ- ent module interacting with syntax on several aspects within the computational system of the language faculty. b) General linguistic properties, inherent to human language, may be represented as features, These features belong to a feature theory module interacting with the grammatical modules, that is morphology, syntax and phonology ©) Inflectional features constitute the morphological expression of some of these features, but not all of them are visible to syntactic operations. According to these assumptions, Case is an abstract notion, a universal notion, belonging to a feature theory module. The encoding of Case takes place within an autonomous morphological module, but Syntax manipulates only the Case informa- 7Angela Ralli tion that is syntactically useful.’ This means that, in languages, there are instances where a word can be morphologically marked for a particular Case value without any need for syntax to provide a syntactic licensing (e.g,, checking) for this value. The assumptions above also predict that all members of the module of features are not overtly realized, i.e, morphologically expressed, in every single language and each grammatical component chooses the features that are appropriate for its own purposes. Since it depends on the particular language to choose the features for its own morphological system, the features of Case, Number and Gender, which are overtly realized in the Greek inflection, belong to the morphological make-up of Greek words, but some of these features could eventually be absent from the mor- phological system of another language, ¢.g., Mohawk. The three features, however, as is further predicted, do not have to be equally visible to Syntax.° Let’s proceed now by examining first the operation of feature coindexation 2.1, Feature coindexation In this section, we would like to propose that coindexation between dislocated con- stituents and pronominal elements affects syntax only because it is a function of the Conceptual) Intentional) level and, following Chomsky 1995, there is no difference between the pre-Spell-Out syntactic operations and those of the C-I Interface. Thus, coindexation should involve semantically relevant features e.g., Number, and exclude grammatical features such as Case. Contrary to Baker’s (1996) claims, we do not be- lieve that this coindexation depends on the overt morphological make-up of the dislo- cated constituents and on the overt expression of features participating in word- formation procedures. Were we to follow Baker’s approach, the Case mismatch ob- served above would remain unexplained since Case overtly participates in the word- formation of Greek nominals (each Case value corresponds to a particular morpheme). Within the minimalist framework (cf. Chomsky 1995), there is a postulation of a [interpretable] feature that has no morphological motivation, but is relevant to syn- tactic purposes, in the sense that the [-interpretable] value designs categories used within the strict limits of syntax while the positive value refers to categories that may also have a scmantic relevance. On the basis of this feature, Case is syntacti- * That Case is an abstract universal notion from which one can deduce concrete forms in Mor- phology as well as more or less concrete notions in the other grammatical modules, has been firstly proposed by Hjelmslev (1935: 85). He also states that these notions cannot be deter- ‘mined in isolation, but only throughout a system of syntactically and semantically Variants. ‘As may be expected by such assumptions, the overt manifestation of Cases is not isomorphic with the set of functions that these Cases may express. For instance, more than one function are expressed by Genitive case in Greek, ¢.g., indirect object, adnominal relation, 68Feature mismatches in Greek clitic left-dislocation constructions cally characterized as [-interpretable], while Gender and Number are [+interpretable].” A [-interpretable] feature, i.e., Case, is invisible to the C-I level, as opposed to the {+interpretable] features of Gender and Number which belong to the features visible to this level. Since coindexation is primarily a C-I level operation, we can explain why coindexation with respect to Case does not oceur between the dislocated constituent and the pronominal element in Greck while, as shown by the ungrammaticality of the data below, coindexation applies to Gender and Number: (9) a. *Opja m’agapai, ton agapo whoever-FEM-NOM me-ACC-loves, _him-MASC-ACC love-1S “I love whoever loves me” (or Whoever loves me, I love him) b. #0 fititis, i kathijites tus agapane olus the student-NOMS, the professors-NOM them-ACC-PL love-3PL all- ACC-PL “The professors love all the students” (or The students, the professors love them alll) Now we can use this distinction of [interpretable] features to modify Baker's fourth condition on non-movement chain formation given in (1), as follows: (10) X and Y are non-distinct in (+interpretable] features (i.e., Number, Gender, Person, etc.) (10) excludes [-interpretable] features, such as Case, from the matching condi- tions on the non-movement chain formation. As already said, the exclusion of [-interpretable} features from the coindexation procedure makes sense if we view the coindexation relation between a dislocated constituent and its clitic as a reflex of a C-I relation, which, by definition, can only be expressed in terms of features inter- pretable at that interface. Notice that Case as well as Number and Gender are also involved in the word- formation procedure of both the dislocated constituent and the clitic. However, despite the fact that the three features receive a separate treatment in Syntax, they behave dif Notice that this [interpretable] opposition is only applicable to the most structural cases, that is to Cases, such as Accusative, that are closely related to grammatical relations. Even these Cases, however, may sometimes be related to a semantic interpretation, de- pending on the particular language, as the following Latin example seems to denote: (ii) Vado Romam Go-IP-S Rome-ACC (destination) “I go to Rome” As Blake (1994: 32) notes. the rationale for separating meaningand functions snot so explicit, 69Angela Ralli ferently in Morphology. For instance, Case and Number belong to the same inflec- tional cluster representing the nominal ending, while Gender characterizes the stem. On the other hand, in clitics and determiners, the same portmanteau morpheme hosts all three features. This difference in use of the same features by Morphology and Syn- tax may be a further proof that the two modules are independent and that each module provides its own means to manipulate the features appropriate for its purposes. Having said this, the following questions remain with no answer, however. a) Why is nominative the only altemative Case value for CLLD constituents, as shown in (I 1a-b)? (1) ao Pavlos, tu pirane to pedi the-NOM Paul-NOM, him-GEN took-away-3PL the kid b. Tu Paviuy tu pirane to pedi the-GEN Paul-GEN, him-GEN took-away-3PL the kid “They took the kid away from Pau!” c.*Ton — Pavlo, tu pirane to pedi the-ACC Paul-ACC, him-GEN took-away-3PL the kid b) Why is “strict identity” of features, i.e. (12b) and (13b), in which the dislocated constituent agrees in Number, Gender and Case with the clitic, better than “partial identity” of features, ie, (12a) and (13a), where the dislocated constituent and the clitic do not agree in Case? (12) 20 Pavlos, tu pirane to pedi the-NOM Paul-NOM, him-GEN took-away-3PL the kid b.Tu = Pavia, tu pirane to pedi the-GEN Paul-GEN, him-GEN took-away-3PL the kid “They took the kid away from Paul” See Ralli (1994) for more detaits on this claim and for an analysis of the Gender feature as an inherent marker of Greek stems. ‘As shown by the following example, the sentence becomes ungrammatical if the dislo- cated constituent appears also in Vocative. (iiiy*Pavle, ta pirane to pedi Paul-VOC, him-GEN took-away-3PL the kid Vocative is the fourth value of the Greck Case system, but it won’t be considered here because of its special character: it is the case form of “address” and marks constituents that stand outside construction, bearing no relation of dependents to heads. On the other hand, as shown by Kurylowicz (196 188) Nominative, Accusative and Genitive are the “most” grammatical cases, that is the cases closely related to grammatical relations, 70Feature mismatches in Greek clitic left-dislocation constructions (13) a1 fitites, i kathijites tus agapane olus the students-NOM, the professors-NOM. them-ACC tove-3PL all-ACC “The protessors love al! the students” (or The students, the professors love them all) b. Tus fitites, i kathijites tus agapane olus the students-ACC, the professors-NOM them-ACC love-3PL all-ACC “The professors love all the students” (or The students, the professors love them all) 2.2, Nominative as a default Case value With respect to the first question, we would like to claim that Nominative acts as a default Case value in CLLD contexts. Since non-movement chain formation does not involve identity of [-interpretable] features, as claimed in (10), any Case value could be possible in CLLD contexts. On the other hand, CLLD contexts do not in- volve Case- checking configurations for the dislocated constituent, which in princi- ple exclude checking of any Case feature on a CLLD constituent, and thus, the oc- currence of any Case marked constituent as CLLD constituent on syntactic grounds, as opposed to what we have seen in Greek. The data we have considered so far show that the dislocated constituent may have either the same Case value as the one on the clitic, or Nominative (cf. (12) and (13). Our proposal then is that Nominative is firstly assigned within Morphology and its presence in Syntax can be considered as that of a default Case value which becomes possible for two reasons: a) dislocated constituents and clitics do not need to agree in Case (cf. (10), and b) all Greek nominals must bear a Case value, as a consequence of their morpho- logical make-up. (b) leads to the possibility for the presence of a default value in nominals for which no particular value is syntactically justified, There are many pieces of independent evi- dence which square quite well with the idea that Nominative should be considered as, the default option in the Greek Case system. For instance, Nominative has atways been the Case outside construction in Greek, the Case for “naming”, (onomazo in Ancient Greek). According to Humbert (1960: 249), it can be conceived independently of any grammatical/syntactic relations. On the other hand, Jakobson (1958) attributes the close relation between Nominative and the concept of topichood (observation which goes back at least to Aristotle) to the fact that Nominative is the Case value with less nAngela Ralli relational content.'° Thus we are allowed to claim that in 2 non-checking syntactic configuration, such as the one involving feft-dislocated constituents, Nominative could appear as the only alternative value expressed by these constituents, beside Genitive (12b) or Accusative (136), depending on the case The proposal that CLLD conte \ts constitute a kind of non-checking syntactic configuration, where Nominative may be used as a default option is also sup- ported by the examples below: it is shown that when there is a clearcut checking requirement for the presence of another case value, i.¢., Genitive or Accusative, Nominative is not possible. (14) a. Opjon u dosis toonoma mu, tha _ ton voithiso whoever-ACChim-GEN give-PERF-2S the name my-GEN, FUT him-ACC help-PERF-1S “Tl help whoever you give my name to” b. Opju n dosis to onoma mu, tha ton voithiso whoever-GEN him-GEN ... ... him-ACC...... c. Opjos tu dosis to onoma mu, tha ton voithil whoever-NOM him-GEN........... .. him-ACC.. da tu dosis to onoma mu, tha _voithiso OM him-GEN........... .. FUT help-PERF-1S *Opjos dosis to onoma mu, tha _voithiso whoever-NOM .. .. FUT help-PERF-1S ‘As shown in (14a), Greek free relative clauses can also contain clities agreeing with the relativized constituent, (cf. Horrocks & Stavrou 1987). With two clities, in the free relative and in the main clause, we have three possible Cases for the relative pronoun: Accusative (14a), Genitive (14b), and also Nominative (14c), although somehow marginally. However, Nominative for the relative pronoun is completely excluded if one of the clitics is missing, as shown by the ungrammaticality of (14d- ¢). This shows that Nominative is only possible on the relative pronoun when the pronoun is coindexed with clities, i.e. ‘a and ton in (14). When one of the clities is missing, which tantamounts to say that there is a movement chain (which requires a Case corresponding to the one checked against the verb without a clitic), then Nomi- native is not an option anymore. Thus we can conclude that Nominative is a sort of '© The idea that the default use of Nominative is related to the notion of topichood is also exploited by Alexiadou & Varlokosta (1996). In their paper, the authors characterize the left-dislocated constituents as instances of “hanging topics” marked for a default Nomina- tive value (1996: 20). 2Feature mismatches in Greek clitic lefi-dislocation constructions sort of default option, only available when the relative pronoun does not check any Case, neither in the free relative nor in the main clause, ‘Assuming now that a default option may equal what can be considered as the unmarked one, this default use of Nominative for syntactic reasons does not coin- cide with the morpholoxially unmarked Case realization, since Nominative in Greek very often constuutes the marked form of the nominal paradigm. Consider, for instance, the very common masculine nouns in -is or -as (ef. below), where the Nominative singular is expressed by the affix -s whereas the Genitive is not overtly expressed (or expressed by the 0 affix). This is another example in favor of the in- dependence of the morphological module: (15) Nominative. Genitive fititi-s fititi “student” tamia-s tamia “cashier” 2.3. Consequences for the Morphology-Syntax Interface The use of a default nominative option in the constructions examined above leads to the consequence that Syntax may actively use only partial morphological informa- tion in its operations. Our analysis crucially shows that, in specific contexts, Syntax ignores pieces of information coming from fully inflected items, i., Case, while ma- nipulates some other information characterizing the same items, ie., Gender, and Number. Until now, we have restricted our attention to cases that do not involve checking configurations, The Greek language, however, also show instances of checking con- figurations where Syntax shows inconsistencies with respect to Case information coming from Morphology. Consider the foliowing pair of examples (16) a. Opjos den simbatho, den tha perasi_ tis eksetasis whoever-NOM not like-IP-S not pass-FUT-38 the exams “Whoever I don’t like, Won't pass the exams” b. Opjon troi poli, ton.—_—koroidevume whoever-ACC eat-3S very much, him-ACC make fun-1PL “Whoever cats very much, we are making fun of him” Given the examples above, a Nominative is checked in a position requiring Ac- cusative (cf, 16a inside the free relative), or an Accusative is checked in a position requiring Nominative (cf. 162 inside the free relative). This does not occur, how- ever, when the verb of the free relative requires Genitive case: (17) a. Opju dosume to vravio, tha jini diasimos whoever-GEN give-PERF-IPL the prize, FUT become-PERF-38 famous “Whoever we give the prize will become famous”Angela Rall b. *Opjos dosume to vravio, tha jini diasimos whoever-NOM give-PERF-IPL the prize, FUT become-PERF-3S famous ©. *Opjon dosume to vravio, tha jini diasimos whoever-ACC give-PERF-IPL the prize, FUT become-PERF-3S famous In (17), only Genitive is the possible Case value for the héad of the free relative, as shown by the ungrammaticality of (17b) and (1 7c), where neither Nominative nor ‘Accusative can check Genitive inside the free relative. This different behaviour of Case values with respect to checking suggests that there is a sort of opposition be- tween Nominative and Accusative un one hand and Genitive on the other, and that this opposition scems to interfere with the operations of checking. In fact, we would like to propose that a special morphological feature [+Genitive] that crosscuts the Greek morphological system of Case values interferes with the syntactic operation of checking as an instance of Morphology-Syntax interaction. The syntactic visibil- ity of this morphological feature makes possible for a constituent bearing an overt Accusative, or an overt Nominative, i.e., a constituent marked as [-Genitive], to be checked in a syntactic context requiring one of the two values (cf. 16). As opposed to this, in a syntactic context requiring [+Genitive], only constituents morphologi- cally marked for Genitive can be checked (ef. 17). The idea of grouping together Nominative and Accusative, against Genitive, is supported from several points of view. Firstly, this division has always been made by traditional grammarians on both syntactic and morphological grounds, and, as noted by Humbert (1960: 248), the history of Greek has always shown a direct rela- tion between Nominative and Accusative. On the basis of head-dependent relations, Genitive is considered to be an adnominal Case as opposed to the adverbial Cases of Nominative and Accusative.'' On the other hand, in the evolution of the language, there are many instances of nouns which have adapted their irregular nominative form analogically to the “more regular” accusative form: (18) Fifth century B.C. Third century A.C. (cf. Browning 1969) Nominative Accusative Nominative a. pater patera --> pateras “father” b. meter metera > metera “mother” (On the basis of syntactic grounds, the same division of cases is also proposed by Simon of Dacia for the Case system of Latin, Simon of Dacia divides the cases in these which express a substance to substance relation and those which do not Gy) Nom, Ace. Gen. Substance to substance ~ - + (€f. Blake 1994: 37). 14Feature mismatches in Greek clitic left-dislocation constructions Case syncretism in Greek also functions on the basis of this opposition. Neuter nouns show the same inflected form in both Nominative and Accusative while their Genitive form is different. Consider, for example, the inflectional paradigm of the neuter noun soma “body” below: (19) Singular Plural Nom. soma somata Ace. soma somata Gen. somatos —_ somaton Moreover, some derived words today, e.g., the diminutives in -aki, do not have an inflected form for Genitive in both Singular and Plural, while they are fully in- flected as far as Nominative and Accusative are concemed: (20) a. pedaki <_ ped- + -aki “small child” child —_small-NOM/ACC-SING. b. pedakia <_ped- + -akia child small-NOM/ACC-PL c. *pedakiu << ped- + -aki child small-GEN-SING d. *pedakion << ped- + -akion child small-GEN-PL The idea of Case checking being influenced by the language-dependent morpho- logical feature of [+Genitive] allows us to explain the situation given in (16) and (17). The question is, however, whether a word bearing a morphologically overt Accusative, or an overt Nominative, can freely check one or the other value. In other words, can Nominative forms alternate with Accusative forms in all appropri- ate contexts? As (21) shows, this is not possible: (21) a. *Ton Jani —efere—_fuludja the-ACC John-ACC brought flowers-ACC-PL b.O Janis efere—luludja the-NOM_John-NOM brought flowers-ACC-PL “John brought flowers” In (la), the DP ton Jani, is morphologically marked for Accusative, but the sentence is ungrammatical because only Nominative is the syntactically checked case for subjects. At this point, we would like to claim that there is a M(orhological) M(odule) Constraint according to which information provided by morphological forms should be transparent as much as possible to syntactic operations (22) *Opacity on Syntactic Operations 75Angela Ralli ‘The constraint in (22) captures the observation that unexpected Case values only raise whenever two syntactic relations mediating Case require different Case values on a concrete lexical item, The MM Constraint in (22) introduces the notion that Case-marked morphological representations are “optimal” expressions of syntactic Case-checking requirements. This allows for an account of why both Nominative and Accusative Case values are possible in (23), but not in (21). (23) Opjos/opjon den simbatho, den tha perasi_ tis eksetasis whoever-NOM/ACC not like-IS__ not FUT-pass-3S the exams “Whoever I don’t like, won't pass the the exams” Following what we have said till now, the sentence in (21a) is acceptable from the morphological point of view, since the [Genitive] feature that interferes with checking simply requires a (-Genitive] constituent form. However, the sentence vio- lates the constraint given in (22) because the form of the DP ‘on Jani, overtly ex- presses an Accusative and does not correspond to the Nominative Case that the word has checked in syntax. Therefore the construction crashes. The alternative form O Janis (c£.21b), marked for Nominative, is thus preferred since the constraint in (22) forces the most exhaustive possible Case matching, between MM forms and Syntax, ruling out Accusative, and “imposiny” Nominative ‘Assuming that the MM Constraint proposed in (22) applies to all syntactic repre- sentations entering MM, a further question arises: why a sentence, like the one given in (23), is acceptable? In (23), the Nominative form opjos “whoever” is morphologically opaque with respect to the Accusative Case that is checked inside the free relative by the verb simbatho “like”: it does not constitute the more “transparent” representation with respect to the checked Nominative. On the other hand, the Accusative form opjon is licenced inside the relative, but it participates in a chain with a pro in Nominative .c., the subject of the matrix verb): from the morphological point of view, it is also opaque with respect to the chain since it does not constitute the more “transparent” representation with respect to the coindexation. Under both choices, Nominative or Accusative, the constraint in (22) is violated once. In contexts such as (23), the two possibilities are equally good candidates, and thus the two are possible morphologi- cal Case values for the relative pronoun whose stem is op/- 2.4, Strict identity of features Before concluding, let us try to give an answer to the last question posed above, that is why strict identity of features is better than “partial identity”, as shown by the examples (12) and (13), repeated here for convenience as (24) and (25): 76Feature mismatches in Greek clitic left-dislocation constructions (24) 0 Pavlos, tw pirane to pedi the-NOM Paul-NOM, him-GEN took-away-3PL the kid bTu Pav, te pirane to pedi the-GEN Paul-GEN, him-GEN took-away-3PL the kid “They took the kid away from Paul” (25) a. 1 fitites, i kathijites tus agapane ous the students-NOM, the professors-NOM them-ACC love-3PL all-ACC “The professors love all the students” (or The students, the professors love them all) b. Tus _fitites, i kathijites tus agapane — olus the students-ACC, the professors-NOM them-ACC love-3PL _all-ACC “The professors love all the students” (or The students, the professors love them all) As proposed above, in a non-movement chain, there is a coindexation between the clitic and the left-dislocated constituent as far as the [+interpretable] features are concemed, ie., Number and Gender. Since [-interpretable] features, such as Case, are not coindexed, the left-dislocated constituent may also appear in a default Nomi- native case value. In (24) and (25), both a. and b. are acceptable. However, it is true that speakers of Greek show a certain preference for the b. form. Since there is no syntactic motivation for this preference (cf. above), we would like to propose that this preference is also due to an intervention of Morphology. Morphologically, Case, Number, and, in some cases, Gender (e.g., in determiners and clitics) are parts of the same inflectional cluster (see section 2.1.). Greek nouns (and DPs) are obliga torily marked for case. Consequently, this preference towards b. could be dictated by the MM Constraint in (22) which requires the most possible matching of Case of the dislocated element to that of the clitic. On the other hand, this application of the MM Constraint is overruled when the default Nominative value occurs, because the use of Nominative in dislocated constituents may be related to interpretative proper- ties (see Nominative related to topichood, as proposed by Jakobson 1958) and, it may be the case that the relation between Morphology and Semantics obviates Syn- tax, something which goes beyond the scope of this paper. 3. Conclusions In this paper, we have shown that an attempt to link the Case forms directly to syn- tactic operations is unwieldy since grammatical relations need not be in a one-for-one correspondence with Case forms, e.g. Nominative forms in CLDC. We argued that Case marking is handled within Morphology and Syntax manipulates only partial Case information coming from Morphology. Our analysis provides a strong argument for 77Angela Ralli the existence of a morphological module, but also a confirmation for the interaction between Morphology and Syntax. The cases we considered here are instances in which either Syntax does not use all the information provided by Morphology, or the syntac- tic operations are affected by constraints coming from Morphology. References Alexiadou, A. & S, Varlokosta (1995). “The Syntactic Relevance of Free Relatives in Modem Greek”, Ms. FAS Berlin and UPenn, Aronoff, M. (1994), Morphology by Itself’ Cambridge, Mass.: MIT Press Baker, M. (1996). The Polysynthesis Parameter. Ox ford/New York: Oxford University Press. Blake, B. (1994), Case, Cambridge: Cambridge University Press. Booij, G. (1997), “Allomorphy and the Autonomy of Morphology”. Folia Linguistica XXXI/1-2, 25-56. Catsimali, G. (1990). Case in Modern Greek: Implications for Clause Structure. Ph, D. Diss. University of Reading. Cheila-Markopoutou, D. (1991). “Problems of Diachronic Syntax: Free Relatives in Medieval and Modem Greek”. Glassologia °-10, 1-30. Chomsky, N. (1995). The Minimalist Proxsram Cambridge, Mass.: MIT Press, Cinque, G. (1990). Types of A-bar Dependencies. Cambridge Mass.: MIT Press. Di Sciullo, A.M, (1996). “Atomucity and Relatedness in Configurational Morphology”. In A.M. Di Sciullo (ed.) Confizuations Essays on Structure and Interpretation, \7-41. Somerville: Cascadilla Press, Hielmslev, L935) La categorie des cas: essais de grammaire generale I. Copenhagen: Munk- sgaard, Horrocks, G. & M. Stavrou (1987). “Bounding Theory and Greck Syntax: Evidence for Wh- movement in NP”, Journal of Linguistics 23, 79-108, Humbert, J, (1960). Syntaxe Grecque. Paris: Klincksieck. Jakobson, R. (1958/1971). “Morphological Inquiry for Slavic Dectension: Structure of Russian Case forms”. Text in Russian, summary in English in 1971 version, In R. Jakobson, Se lected Writings il: Words and Language, |S4-183. The Hague: Mouton. Kurylowicz, J. (1964). The inflectional Categories of Indo-European. Heidelberg: Carl Winter Phitippaki-Warburton, 1. (1990). “The Analysis of the Verbal Complex in Modem Greek”. Studies in Greek Linguistics. Thessaloniki: Kyriakides. Philippaki-Warburton, 1. & M, Stavrou (1986). “Free Relatives in Modem Greek”. Saedies in Greek Linguistics. Thessaloniki: Kyriakides. Ralli, A (1994), “Feature Representations and Feature-passing Operations in Greek Nominal In- flection”, Proceedings of the 8th Symposium on English and Greek Linguistics. Thessalo~ niki: English Dept. Aristotle University of Thessaloniki. 7Feature mismatches in Greek clitic left-dislocation constructions ~~~ (1997). “Inflectional Features and the Morphological Component Hypothesis”. Paper pre- sented at 20th GLOW. Rabat. To appear in Working Papers of Greek and English Lin- _guistics. Thessaloniki: English Dept. University of Thessaloniki, (1998). “On the Morphological Status of Inflectional Features: Evidence from Modem Greek”. In G. Horrocks, B. Joseph and 1. Philippaki-Warburton (ed.) Themes of Greek Linguistics Il. Amsterdam: John Benjamins. Tsimpli, i. (1990). “The Clause Structure and Word Order in Greek”. In UCL Working Papers in Linguistics. ‘Tzartzanos, A. (1946). Neoelliniki Syntaxis tis Koinis Neoellinikis (Greek Syntax of Modem Greck Koine). Athens: OEDB.BILINGUISMO PRECOCE E ACQUISIZIONE DELLE PERIFRASI VERBALI E DEL CRITERIO WH- IN ITALIANO Franco BENUCCI Questo contributo* nasee al seguito di un precedente studio (Benucei (1997)) dedicato alla sintassi dei prefissi d’aspetto nelle perifrasi verbali dell’italiano e di altre fingue romanze, in connessione con altri noti fenomeni sintattici che oecorrono negli stessi contesti (Clitic climbing, Cambio d’ausiliare, ecc.), generalmente sus- sunti sotto etichette come Clause union o ‘Ristrutturazione’. Focalizzeremo qui Pattenzione su alcuni aspetti della sintassi delle perifrasi verbali, segnatamente sul comportamento del prefisso aspettuale di ripetizione (ri-) e sulla sua interazione con i contesti wh-, nelVinterlingua di acquisizione dell’italiano in una situazione di bi- linguismo precoce italiano-ungherese, formulando precise ipotesi esplicative per i peculiari fenomeni osservati ¢ argomentando cosi in favore dell’esistenza di feno- meni sintattici analizzabili in termini di Long Head Movement. Avremo inoltre oc- casione di osservare significativi paralleli tra lo sviluppo ontogenetico dell’ italiano ¢ quello diacronico della pits ampia famiglia linguistica romanza e di muovere alcuni passi nel campo del bilinguismo, esaminando (¢ respingendo) !"eventualita che al- cuni dei fenomeni analizzati siano dovuti ad una qualche forma di interferenza sin- tattica dell’ ungherese e portando quindi nuovi argomenti contro l’ipotesi che vede nell’acquisizione del bilinguismo precoce I’esistenza iniziale di un unico sistema linguistico in progressiva differenziazione, ¢ a favore invece dell’ ipotesi contraria, di aequisizione del bilinguismo secondo sistemi distinti ¢ diversamente ritmati per ogni lingua. * Ringrazio Gianpaolo Salvi, Cecilia Poletto, Szilvia Papp, Alessandra Tomaselli ¢ il ‘pub- blico’ del XXIV Incontro di Grammatica Generativa e del Seminario di Ricerca in Roma- histica dell"Universita ELTE di Budapest per gli utili commenti e osservazioni alle ver- sioni preliminari di questo lavoro. La responsabilita di quanto sostenuto rimane natural- mente mia. 81Franco Benucci Il corpus di osservazione primario é costituito da enunciati di parlito spontaneo, rilevati nel corso della normale interazione familiare, di mio figlio primogenito Ar- pad, un bambino, all’epoca, di circa 4 anni (dai 42 ai $4 mesi, periodo febbraro 1997 - febbraio 1998), bilingue italiano-ungherese, in una fase di avanzata acquisizione di entrambe fe lingue ¢ di quasi perfetta padronanza delle loro corrispondenze lessicali, sintattiche, ece. (capacita di traduzione estemporanea tra membri dei diversi nuclei familiari, conversazioni familiari mistilingui, capacita di ingenua analisi morfolessi- cale ed ‘etimologica’ di nomi propti,! traduzioni sintatticamente corrette in una lin- gua di espressioni idiomatiche dell’altra,2 ecc.). Mi riferisco con questo alle spiegazioni cercate o proposte da Arpad per i cognomi dei propri compagni di scuola (materna) ¢ degli intestatari di alcune vie padovane, ad cs. Carletto Giratello (in realti Girardello) perché gira sempre. Emanuele Peretta perché mangia sempre le pere. Pits sottle, e dircttamente pertinente per il bilinguismo, é, di fron- te al mio annuncio Oggi andiamo in Via Viacovich (Gianpaolo, anatomista, 1825-99), la domanda Cosa vuol dire Via? avendo “riconosciuto’ nella seconda:parte del cognome (certo udito per la prima volta) ungherese kavics [*kavitl] ‘sasso’ (episodio della prima- vera 1997). Queste osservavioni'e le successive, lungi dal voler indulgere alla celebrazio- ne affettiva delle abilita linguistiche di Acpad, sembrano direttamente rifevanti per inqua- drare la situazione del suo ambiente linguistico e del grado di sviluppo cognitivo che esso Pud aver indotto: episodi come quelli sopra riportati ci sembrano infatti portare conferma alle ipotesi avanzate fin dal 1962 da E. Peal e W. Lambert, secondo cui ’educazione bi- Jingue svituppa nei bambini una precoce consapevolezza dell"arbitrarieta del linguaggio e uuna notevole capacita di astrazione, inducendoli “pid precocemente ed intensamente dei monolingui a riflettere sul funzionamento del linguaggio e a fare delle considerazioni a livello morfologico e sintattico” (cfr. Carpene (1995-96: 9, 38-9)). Degno di nota sembra anche il fatto che tali spontanee ‘considerazioni” morfosintattiche hanno preceduto di vari mesi la sensibilita ¢ le esplicite considerazioni di ordine fonologico (sillabazione, ri- ‘me/omoteleuti, riconoscimento di classi fonemiche (Bianca fa rima con Vian), ecc.),.in- dotie, almeno in parte, dall’attvita scolastica 2 Ad esempio, costruzioni ungheresi a ‘copula @" come nekem jé ‘mi va bene, mi piace’ o nem jatek ‘non si gioca cosi’ venivano inizialmente rese in italiano con una traduzione lctterale degli element lessicali ma con l’inserimento della copula mancante (a me & be- ne, non é gioco), con un risultato idiomaticamente errato, ma sintatticamente cortetto © significativo del livello di competenza nelle strutture sintattiche delle due lingue nelle to- 10 cortispondenze ed opposizioni. Analogo fenomeno, ma in direzione contrarta, si ha nelle rese letterali di Mamma, puoi farmi questo? con Anya, megesindlhatod e=1? (verbo con suffisso potenziale-permissivo invece del condizionale megesindlndd o del presente megesindlod) e Jo ero Batman (ne! gioco) con En voltam a Batman (verbo al passato in- vece del futuro /eszeky: in questi casi é il materiale linguistico ungherese che viene utiliz- Zato in modo grammaticalmente corretto per esprimere perd valori semantici e idiomatici propri deli’italiano, 82Bilinguismo precoce e acquisizione delle perifrasi verbali e del criterio wh- in italiano La lingua target dell’acquisizione3 parlata nel ramo paterno della famiglia ¢, in modo statisticamente prevalente, nell'ambiente (asilo, amichetti e relative fa- miglie, varia socialita, ecc.), & un italiano certo colto ma inevitabilmente caratte- rizzato in senso regionale veneto, le cui caratteristiche (qui) rilevanti possono es- sere cosi riassunte: - collocazione standard di clitici e prefissi d’aspetto con i tempi verbali semplici: Jo rimetto, ne riparliamo - collocazione standard di clitici e prefissi d’aspetto con i tempi composti, rispet- tivamente sull’ausiliare e sul participio passato dipendente: !'ho rimesso, ti ho rivisto - assenza quasi totale di fenomeni di Risirutturazione nelle perifrasi modali, con col locazione di clitici e prefissi d’aspetto sull’infinito dipendente ¢ mantenimento delPausiliare di base del verbo modale: voglio rimetterlo, devo rifarlo, ho dovuto andarci. L italiano parlato (con ottima padronanza) dalla madre, ungherese ormai da oltre quindici anni residente tra Mantova ¢ Padova, mantiene invece la connotazione normativa dovuta all’apprendimento scolastico in Ungheria, con prevalenza statisti- ca della collocazione alta dei clitici anche nelle perifrasi modali e Cambio dausiliare: lo voglio rimettere, lo devo rifare, ci sono dovuta andare. Va peraitro sottolineato che la madre parla con i figli quasi esclusivamente in ungherese, riservando italiano alla conversazione adulta ed allinterazione con e- iementi esterni alla famiglia. Avviene invece esclusivamente in ungherese Pinterazione con il ramo materno della famiglia (durante i regolari e prolungati sog- giorni in Ungheria e le frequenti visite a Padova di nonni e zit ungheresi) e, in pas- sato, con una baby-sitter ungherese alla pari: una situazione quindi di bilinguismo ambientale abbastanza bilanciato, con un’ educazione linguistica dei bambini ispirata al principio ‘un genitore ~ una lingua’. Per completare il quadro familiare, accenne- remo al fatto che il fratellino Nandor, all’epoca dei fatti osservati, aveva circa due anni (tra i 17 i 29 mesi) e che fa sua produzione linguistica si trovava ancora lar- gamente in una fase pre- (0 proto-) sintattica. II quadro danalisi in cui svilupperemo lipotesi € quello delineato in Benucci (1997), cui rinviamo per i dettagli, e si basa sullo schema di struttura funzionale del- la frase di Cinque (1997: 4.30) integrato con un’analisi ‘alla Kayne’ della fenome- nologia di piazzamento dej clitici, In tale quadro, i clitici pronominali ed i prefissi @aspetto corrispondono a diverse teste della struttura funzionale: nel nostro caso 3. © almeno presuntivamente tale: come vedremo in chiusura, la fase (per ora) finale dello sviluppo detle perifrasi modali di Arpad é per certi versi pi vicina all’italiano materno (Clitic climbing frequentissimo). 83