Dei ex Machina: a note on plural/mass indefinite determiners. Roberto Zamparelli 1 7 August 2002

Dei ex Machina: a note on plural/mass indefinite determiners Roberto Zamparelli 1 7 August 2002 FIRST DRAFT, COMMENTS WELCOME 1 Introduction The structure and meaning of Romance plural/mass indefinite nominals has recently been a matter of some debate (Chierchia 1998a; Storto 2001; Roy 2001). Meaningwise, the determiners that introduce these nominals, degli and des in (1), behave as plural or mass indefinite articles, equivalent to the English some in some people/water. Morphologically, however, degli and des appear to be composed of the preposition di of incorporated with the regular form of the definite article (il/lo/la/i/gli/le in Italian, le/la/les in French; I will henceforth use dei and des for the whole paradigm). (1) a. Ho incontrato [degli studenti]. I have met [of the students]. [lit.] I have met some students. b. J ai rencontré [des étudiants]. I have met [of the students]. [lit.] I have met [some students]. Italian French The process by which the definite article and a preposition combine into a preposizione articolata is completely standard in Italian and French. In particular, the combination of +Def-Art can be found in overt partitives, illustrated in (3) (see Selkirk 1977, Hoeksema 1996). (2) a. Ho incontrato tre degli studenti. I have met three of the students. b. J ai rencontré trois des étudiants. I have met three of the students. The question addressed in this article is whether the similarity between (1) and (2) justifies deriving the former from the latter in a compositional fashion; whether in other terms (1) are partitives without a number. The answer I will propose is partly positive, partly negative: I will argue that dei/des are compositionally derived via a complex structure akin to that of partitives, but that the partitive semantics applies not to a normal definite nominal, but to a kind-denoting 1 Facoltà di Lettere, Università di Bergamo, P.zza Vecchia 8, 24129, Italia roberto@unibg.it. Thanks to Alan Munn for comments and corrections. All remaining errors are my own. 1

definite of the type studied in Zamparelli (2001). Specifically, I propose that the base of a plural/mass indefinite like (3a) is not the definite we find in (3b) and which must refers to some unique or familiar set of dodos, but the one we see in (3c), which would be rendered in English by a bare plural. (3) a. [Dei dodo] dormivano nella mia gabbia. [of the dodos] slept in my cage Some dodos slept in my cage b. [I dodo] stanno dormendo nella gabbia. [the dodos] are sleeping in the cage the dodos are sleeping in my cage (some specific ones). c. [I dodo] si sono estinti [the dodos] self are extinct Dodos have become extinct (as a species) The structure of the paper is as follows. In section 2 I will present the view that dei/des-nominals are partitives. In section 3 some old and new problems for this proposal will be reviewed. Sections 4 and 5 will give the core of my proposal and discuss the role of di. Section 6 address some points related to economy. 2 Chierchia s theory Considering the similarity between (1) and (2) Chierchia (1998a) proposes that the two constructions are structurally identical ((4) and (5)) and refers to the former as the bare partitive construction. (4) a. Dei of the folletti. elves b. [ DP [ D D ¼ ] [ NP [ N 0 Ô ÖØ ] [ PP di [ DP i folletti]]]] c. [ DP [ D dei ] [ NP [ N t ] [ PP t [ DP t folletti]]]] (5) a. Alcuni dei folletti. Some of the elves b. [ DP [ D alcuni] [ NP [ N 0 Ô ÖØ ] [ PP di [ DP i folletti]]]] According to Chierchia, the only difference between bare and overt partitives is the lack of an overt numeral in the higher D position. In order to license this position, the complex definite article + preposition de+i moves to D, the position occupied by the numeral (5c) in overt partitives. In Chierchia s analysis 0 Ô ÖØ is an empty relational noun which selects an obligatorily definite DP and is responsible for the partitive meaning (i.e. 0 Ô ÖØ denotes x y[y x]). The composition between 0 Ô ÖØ and the definite article gives (6a); the next step, raising de+i to D triggers a existential type-shift (i.e. Q x[p(x) Q(x)] composes with the denotation of de+i to yield a generalized quantifier (6b)). (6) a. 0 Ô ÖØ Æ the = P x[ x P] 2

b. [ DP dei ] = Q x[p(x) Q(x)]) Æ P x[ x P] = P Q x[ x P(x) Q(x)] As Chierchia points out, one syntactic advantage of this analysis is the impossibility of coordinating the indefinite determiner dei with other determiners, which becomes an attempt to conjoin a head, the numeral, with a whole phrase, the one containing the chain of dei (7b): (7) a. Uno o *dei / due ragazzi. One or of the / two boys. b. *[ DP [ D uno] o [ DP dei... [ DP t ragazzi]]] Raising the de+i complex to D gives it the scope possibilities of a full-fledged determiner. In particular, Italian dei can take scope over other operators, much as the English plural indefinite determiner some, and unlike bare plurals. (8) a. Non ho visto ragazzi. Not I have seen boys I haven t seen boys only b. Non ho visto dei ragazzi. Not I have seen of the boys Ô I haven t seen some boys or There are boys I haven t seen Interestingly, even though the dei-nominal as a whole has an indefinite meaning, it cannot be used in negative existential statements: (9) cannot mean Elves do not exist, only Some elves are not present (in some spatio-temporal location). (9) Non ci sono dei folletti. not there are of the elves There aren t some elves According to Chierchia, this follows from the fact that unlike a bare plural the bare partitive in (8b) contains a -operator, which triggers the existence presuppositions normally associated with definiteness. The conflict with existential statements is predicted by analyses in which the internal noun phrase of a felicitous existential statement must not carry any presupposition of existence of its own (e.g. Zucchi 1995). Thus, the partitive analysis captures both the otherwise coincidental homophony between an article and the partitive, and a peculiar aspect of their distribution. 3 Storto s criticism Storto (2001) raises various arguments against Chierchia s bare partitive analysis. One concerns cross-linguistic differences between plural/mass indefinite articles in Italian and French. Unlike Italian dei, which can take either wide or narrow scope, French des must have narrowest scope, like Italian and English existentially interpreted bare plurals (10a). Moreover, des-nominal can- 3

not appear as preverbal subject of unaccusative predicates (10b), unlike the corresponding Italian cases (examples from Roy 2001 and Delfitto 1993 respectively, cited in Storto 2001). (10) a. Tout les visiteurs ont lu des journaux all the visitors have read of the newspapers All the visitors have read newspapers b. *?Des linguistes sont arrivés of the linguists are arrived no reading Thus, it seems that, without additional assumptions, no common analysis can be given for the meaning and distribution of bare partitives in Italian and French. Even setting French aside, it turns out that the -operator embedded in the semantics of bare partitives makes predictions that are too strong. First, if the ill-formedness of (9) as a pure existential statement is due to presuppositions of existence associated with dei, positive existential statements should be expected to be just as bad (much as?? there are the elves ), while they are perfectly fine. (11) So che ci sono dei folletti e prima o poi ne troverò. I know that there are of the elves and sooner or later I will find them I know elves exist, and sooner or later I ll find them Indeed, the same pattern is found even in English with the determiner some: (12) a. *There aren t some elves (any more) b. There are (still) some elves. Storto concludes that the problem must lie in the negative polarity of these sentences, and suggest a connection with the fact that dei cannot be focused (see Storto 2001, sec. 4.3). Presuppositions of existence do not play any role in this explanation. A second and more general issue is that Chierchia s analysis predicts synonymy between (13a) and (at least some reading of) (13b), modulo the semantic effect of P+D raising to the upper D. (13b) is in fact like (13a), except for the overt presence of alcuni some ÔÐÙÖ. (13) a. Dei folletti. of the elves. Some elves b. Alcuni dei folletti. some of the elves. Some of the elves Storto (2001) gives (14) (his example (10), see also his sec. 4.3) as a case where this synonymy does not obtain. (14a) is contradictory because the use of a full partitive presupposes the existence of a larger group of martians which have not landed. No contradiction arises in (14b), since 4

here there is no implicature that the group that landed is part of a (specific/definite) larger group. 2 (14) a. #[Alcuni dei marziani che sono atterrati [some of the martians that are landed loro sono gli ultimi della loro specie. they are the last of their species. b. [Dei marziani che sono atterrati nel mio [of the martians that are landed in my sono gli ultimi della loro specie. are the last of their species. nel in mio my giardino] garden] giardino] garden] mi hanno detto told me mi hanno detto told me che that che loro that they Some martians that have landed in my garden told me they are the last of their species An even sharper contrast comes from cases of generic quantification: (15a) is compatible with a situation in which nearly all the genuine Italians do not like spaghetti; this is not true of (15b), as the paraphrasis makes clear. (15) a. Alcuni dei veri italiani mangiano gli spaghetti. some of the true Italians eat the spaghetti. Some of the true Italians eat spaghetti. b. Dei veri italiani mangiano gli spaghetti. of the true Italian eat the spaghetti. True Italians eat spaghetti To be sure, there are contexts in which the -operator embedded in the bare partitive does seem to show up. In (16) dei and alcuni dei are fairly interchangeable (Chierchia, p.c.): (16) Alla festa c At the party there erano molti ragazzi were many boys dei ragazzi sono andati via. of the boys have left. e ragazze. Poi, alcuni dei ragazzi / and girls. Then, some of the boys / Here dei ragazzi can mean some of the boys that have just been mentioned. On further reflection, however, this possibility does not necessarily mean that dei embeds a definite DP, but simply that, like all other plural indefinites (and unlike the singular a ), dei has a D-linked reading available. The first sentence above can in fact be continued with: (17)... Molti / alcuni / tre ragazzi se ne sono andati(, gli altri sono rimasti)... many / some / three boys have left(, the others have stayed) 2 Note that in order to treat (14)b as a bare partitive Chierchia must crucially abandon the constraint of proper partitivity, i.e. the notion that a partitive denotes a plurality smaller that he one denoted by the definite it embeds (see Barker 1998). For dei marziani to mean the total set of martians, the relation expressed by 0 Ô ÖØ must reduce to equality. 5

where the indefinites can be interpreted as many/some/three of the boys there were. Obviously, this meaning cannot be due to the possibility that these indefinites embed a hidden definite determiner. Considering these facts, the presence of an -operator in the semantics provided by the partitive analysis seems to be more of a hindrance than an advantage. The partitive account can stand only in conjunction with a principled account for why the definite inside the bare partitive is not generally felt to be there. Lacking such an account, Storto concludes that the uniform partitive analysis is not viable, that bare partitives are after all not partitives at all, and that Italian dei and French des are lexical determiners, with distinct scopal properties. It seems to me that this solution throws away the baby with the bath water. Still unexplained is why, in various languages, a lexical indefinite determiner should have developed, ex-machina, in the shape of a prepositional phrase. In addition, the Italian/French scope differences must be stipulated. Last but not least, dei/des cannot be reduced to a Romance equivalent of some, for two reasons. First, unlike dei and the singular indefinite article, English some cannot be bound by generic operators or adverbs of quantification. Contrast (15)b with (18): (18) Adolescents / an adolescent / *some adolescent(s) is/are (often) tall. Second, unlike dei, some/alcuni/quelques cannot appear in predicate nominals such as (19). 3 (19) Quei venti uomini sono *alcune / delle brave persone. those 20 men are some / of the good people These differences could perhaps help justify the development of an additional form of indefinite determiner in Romance languages, but they leave open the issue of how these effects arise and whether they can be traced to a single parameter. Let s therefore search further in the neighborhood of the partitive analysis. 3.1 Some advantages of a dei/des -raising analysis Chierchia s analysis can be factored in two parts: the idea that bare dei-nominals are underlyingly partitives, and the idea that their external DP is licensed by raising the determiner and its associate P from the internal DP. Let s call a theory that adopts the second assumption a dei/des-reasing analysis. I want to suggest that a dei/des-raising analysis is better suited than a purely lexical account to deal with the binding and predicative facts in (18) and (19). Consider how a possible analysis would run. As we saw in (15), dei-nominals can be bound by a generic operator, but alcuni some cannot, regardless of the presence of a partitive. (20) Alcuni degli italiani / alcuni italiani amano gli spaghetti. Some of the Italians / some Italians like the spaghetti It cannot mean: Generically speaking, Italians like spaghetti 3 English some has a modifier of amount/extent reading (e.g. Those guys are quite some studs!) which is acceptable in predicative position. I take this meaning to be irrelevant here. 6

It seems plausible that operator binding of D is possible only when this head position is not filled by some over lexical operator (which would preempt the external binder). Suppose, in particular, that when binding of D (a variable) is provided by an external adverbial or by narrowscope existential closure, dei has not raised all the way to D, but has stopped in NumP, a position immediately below D which hosts numerals. Dei trasmits its features to Num, and Num in turn identifies D under predication (see Heycock and Zamparelli 2000 for details of this approach), licensing the DP (21a). (21) a. [ DP D [ NumP dei [ NP t [ PP t [ DP t ragazzi]]]]] b. [ DP [ D dei ] [ NP t [ PP t [ DP t ragazzi]]]] If, on the other hand, dei raises all the way to D as in (21b), it functions as a full-fledged quantifier, and can take wide scope using the same means as other indefinite determiners. English some can now be analysed as a determiner which is base generated in D. As a result, no adverbial binding of some obtains. 4 If, as argued in Zamparelli (1995), predicate nominals are subparts of DPs which exclude the DP layer (e.g. bare NumPs), we capture the fact that some is not able to head predicate nominals: (22) a. Quei venti uomini sono [ NumP delle... [ DP t brave persone]] = (19) b. *Those 20 men are [ DP some good people] The scope differences between French and Italian can now be approached along the same line, and tied to the well-known fact that French does not have bare plurals. Longobardi (1994) has proposed that in languages like English bare plurals are licensed by moving N (or N s features) to D at LF. This option must evidently be disallowed in French; one way to express this is to say that in French the D head cannot be licensed by elements with nominal features (or perhaps plural nominal features ; this formulation would still allow proper names in D). 5 (23) a. [ DP dogs [ NP t ]] English: N-features in D acceptable b. *[ DP chiens [ NP t ]] French: N-features in D not acceptable But now the very same principle rules out the structure (21)b, and only allows (21)a in French: des cannot move to D, since it is by hypothesis endowed with nominal (and prepositional) features. As a result, des stops at NumP. D remains empty, but is licensed under predication by the NumP hosting des; however, since des never reaches D, it never directly binds this position. As a result, D can only receive an existential interpretation via Existential Closure, which gives narrow scope, and it is not allowed in positions which probably cannot be reached by Existential Closure (see (10)b above). 4 Some evidence for a D position of some comes from the fact that this determiner can be coordinated with full-fledged quantifiers such as most, even though, as we have seen, it cannot be coordinated with numerals. (i) We will buy some or most /??twenty (of the) art pieces in this exhibit. 5 I am disregarding the issue of pronouns here. Another apparent exception is the coordination of bare nouns, which according to Heycock and Zamparelli (2000) move to [Spec,DP], not D. 7

3.2 Additional syntactic problems Considering these advantages, it seems worthwhile to try to construct a dei-raising derivation which preserves the insights of Chierchia s bare-partitives but avoids the problems seen to far, plus two serious syntactic pitfalls that Storto doesn t discuss. First, in regular partitives the embedded definite DP may include numerals (24a). Yet, internal numerals are completely impossible in bare partitives (24b). (24) a. Alcuni dei dieci ragazzi. Some of the ten boys. b. *Dei dieci ragazzi. Of the ten boys. Second, bare partitives do not seem capable of licensing a double restrictive relative, while full partitives may do so (Selkirk 1977). (25a), with a full partitive, refers to the following situation: yesterday I have met some boys; today I have met again just some of them; part of this latter group is of Dutch nationality. The same interpretation is impossible in (25b): the two relative are felt to modify the same group of people. Since the modifiers are of exactly the same type, the natural way to compose them is via coordination, as in (25c), hence the non-naturalness of (25b). (25) a. Alcuni dei ragazzi some of the boys oggi sono olandesi. today are Dutch. b.??dei ragazzi che of the boys that olandesi. Dutch. c. Dei ragazzi che of the boys that sono olandesi. are Dutch. che ho incontrato ieri che ho riincontrato that I have met yesterday that I have met again ho incontrato ieri che ho riincontrato oggi sono I have met yesterday that I have met again today are ho incontrato ieri e che ho riincontrato oggi I have met yesterday and that I have met again today Since relative clauses may modify nominals with no overt determiners, it is not immediately clear why the most external relative shouldn t be able to modify the uppermost DP. On the other hand, if Storto is right, the bare partitive is not at all composed of two stacked DPs, but is a single DP which cannot be restricted at different points. This is in fact precisely the argument that led Selkirk to a single-dp analysis of pseudo-partitives such as a number of objections. 8

4 Romance kind-denoting definites As is well known since Contreras (1986) and Casalegno (1987), Romance languages use plural or mass definites in sentences where English would use bare plural count or bare singular mass nominals with a generic or kind reading. The following are representative examples from Italian. (26) a. [I cani] si trovano un po in tutte le taglie. [the dogs] come a bit in all the sizes. Dogs come in all different sizes. b. [I pitbull] sono impopolari in Inghilterra. [the pitbulls] are unpopular in England Pitbulls are unpopular in England. c. [Gli uccelli] si sono evoluti da[i rettili]. [the birds] evolved from[the reptiles]. Birds evolved from reptiles d. [L elio] è abbondante nell universo. [The helium] is abundant in the universe Helium is abundant in the universe. e. [I fisici nucleari] studiano [gli atomi]. [the physicists nuclear] study [the atoms]. Nuclear physicists study atoms f. Pasteur scoprì [gli antibiotici]. Pasteur discovered [the antibiotics] Pasteur discovered antibiotics g. Il Quinto Giorno, a mezzogiorno spaccato, Dio creò [le zebre]. On the fifth day, at 12 o clock sharp, God created [the zebras]. On the fifth day, at 12 o clock sharp, God created zebras. Since Carlson (1977), the predicates in (26) have been taken to select a kind sort in the argument(s) in brackets. Following Zamparelli (2001), I will assume that the Italian bracketed definites above denote kinds of things, much as in Carlson s proposal for bare plurals in English. My proposal is that the partitive determiner dei/des is not a special indefinite determiner, but rather a Preposition+Determiner complex in a raising structure built on a kind-denoting definite. 6 In first approximation, this means adopting the structure proposed in (Chierchia 1998a), repeated below in (27), but with the internal definite DP denoting a kind. This structure will be slightly modified in the next section. (27) [ DP [ D dei ] [ NP t [ PP t [ DP t folletti]]]] 6 This idea was brought to my attention by the journalist Katia De Gennaro; it is in fact not new (see in particular Renzi 1995, pg. 374), but I have never seen it analysed in any detail. Alan Munn (pc) tells me that a similar proposal has been made for Japanese in Kakegawa (2000). 9

Informally, the composition of a bare partitive such as (28a) would not be (28b), but (28c). (28) a. Ho incontrato degli studenti. =(1) b. I met some of [the unique group of students in context] Base: Individual definite meaning c. I met some instances of [the student-kind] Base: Kind definite meaning Consider two immediate advantages of this proposal. One property of Romance kind-denoting definites is that their extensional existence presuppositions are void. There is nothing contradictory or redundant in (29): (29) I fantasmi non ci sono, (mentre i vampiri esistono). the ghosts not there are, (while the vampires exist) There aren t any ghosts, (but vampires exist) According to this statement, there are no instances of the ghost-kind in the current world/time (kinds must have realizations in some possible world, but not necessarily in the actual one). This automatically explains the vanishing existence presuppositions of dei-nominals. A second relevant property of Romance definites is that the kind reading is blocked if a numeral is inserted after the article (see Dayal 2000 for an account). For instance, even if there is a total of 3000 wolves on Earth, (30a) can only be a odd statement about a particular group of wolves the normal individual-level interpretation of the definite is forced. (30) a.??[i tremila lupi] si sono evoluti molto lentamente. [the 3000 wolves] evolved very slowly b.??[i tremila lupi] diventano più grandi come si [the 3000 wolves] become bigger as one viaggia travels a nord. North A consequence is that the non-existent form *dei 3000 lupi of the 3000 wolves cannot be derived. It cannot come from the structure (31a) (pre-movement), which should be a candidate if the internal DP were a normal definite; and it cannot originate from (31b), with the numeral based in the external DP. This is excluded (also in the original partitive analysis) by the fact that the de+i complex would have to hop over or stop in the head position occupied by the numeral. (31) a. *[ DP... [ PP de [ DP i 3000 lupi]]] b. *[ DP dei [ NumP 3000 [ PP t [ DP t lupi]]]] A serious potential objection to a kind-based analysis is that not all definite nominals can be interpreted as kind-denoting definites. In particular, definites like those in (32), with modifiers anchored to rigid designators, non restrictive prenominal adjectives or prenominal possessives, cannot normally be interpreted as kinds-of-things (see Zamparelli 2001, sec. 4.3): 7 7 More precisely, a definite with a non-restrictive adjective can be well-formed with a kind-denoting predicate if it can be interpreted as a normal definite anaphoric to a kind which is contextually salient. The same is of course true in English: Pittbulls are common nowaday... [These dogs] evolved from sausages.. Anaphoric links are unnecessary for genuine kind-denoting definites. 10

(32) [I cani che erano qui ieri] / [I miei cani] / [I simpatici cagnolini] [the dogs that were here yesterday] / [the my dogs] / the cute doggies] (??sono ( are estinti). extinct) (I am excluding the subkind or taxonomic interpretation here). Yet, the same definites can easily appear with dei. (33) a. C erano dei simpatici cagnolini. there were of the nice doggies. b. Ho visto dei cani che erano qui ieri. I have seen of the dogs that were here yesterday. A first reaction to these data might be to retreat to the idea that bare partitives are ambiguous: they can embed either normal definites or kind-denoting definites. The latter would have no existence presuppositions, the former would. (33) would now be a case where the first option is forced. This optionality, however, is neither theoretically desirable nor empirically correct: it undermines the explanation for the absence of dei 3000 cani of the 3000 dogs, which would now have a possible derivation. Moreover, the dei-nominals in (33), which are by hypothesis based on normal definites, should always carry presuppositions of extensional existence. This happens to be true for (33b), for independent reasons, 8 but is definitely false for (33a). Fortunately, there is another way to accommodate the data in (33). Note that the Romance definites which could not be interpreted as kinds, in (32), always had a modifier of some sort. Nothing forces us to assume that the modifiers we see in dei-nominals always attach inside the kind-denoting definite. Consider an alternative derivation: (34) a. [ DP½ dei simpatici... [ PP t [ DP¾ t cagnolini]]] b. [ DP½ dei [ PP t [ DP¾ t cani]] che erano qui ieri] In (34a), the non-restrictive adjective (in a Spec or Adjunct position, see Cinque 1994) is attached in the outer DP, and is passed by the de+i complex on its way to D/Num. In (34b), the relative is attached at some level in DP ½, not DP ¾. Since outside the PP the denotation is no longer a kind, any extensional modifier can combine with it without any problem. 9 This possibility also sheds some light on the problem of double modifiers in example (25). In (25)b, neither modifer can modify a kind, hence both are forced to apply to the external DP, giving the illusion of a single DP. In fact, when the internal modifer is compatible with a kind, there is no problem with stacking: (35) [De[gli animali con le orecchie lunghe] che ho visto oggi] [of [the animals with the ears long] that I have seen today] 8 The presuppositions come from the relative, not the article; cf. a dog that I met here yesterday, which does presuppose that dogs exist. 9 In a head-internal analyses of relatives, the whole PP-DP ¾ complex would have been moved from inside the relative. 11

5 The nature of di Assuming that the internal DP ¾ denotes a kind, we turn to the role of di. In Chierchia s account, di is meaningless, and the semantics comes mostly from a null partitive N. In the present account the partitive N is superfluous, so the semantic burden must be shifted back onto P. There are in fact other Italian examples where we see di with kind-denoting nominals (DPs such as a strange breed / a rare color), for instance the following predicational structures, where di is obligatory: (36) Fido era [??(di) una razza nana / un colore strano ] Fido was (of) a breed dwarf / a color strange Fido was (an animal) of a dwarf breed / a strange color One obvious approach is to say that the preposition that introduces dei-nominals is the lexical realization of the type/sort-shifter, the up operator defined in Chierchia (1998b), which maps kinds into the set of their instanciations: (37) [ of DP] = DP = x x is an individual instanciation of DP with DP kind-denoting Apart from the proliferation of di-meanings which results, this idea is problematic with respect to the reasonable proposal that languages cannot employ hidden type-shifting operators if they have overt lexical elements with the same meaning (see Chierchia 1998b, Dayal 2000). If this view is on the right track, a di meaning would end up blocking the application of in all di-less contexts probably an incorrect prediction. One alternative is to adopt the system in Zamparelli (1998), which tries to give a unified account of partitives and constructions like (36) above, and extend it to the case at issue. The basic idea of this system is the following (here I simplify the internal DP structure). The di/of of partitives is an operator with two arguments, one in complement position and the other in specifier position. This operator, called R ( residue ) returns the denotation of its specifier minus the denotation of its complement. The complement is a full DP. The specifier is filled by a copy of the NP embedded in the complement DP. At spell-out, one of the copies of this NP (usually but not always the upper one) is not pronounced. This gives the following syntax and semantics: (38) a. [ RP [ NP boys] [ R ¼ of [ DP the [ NP boys] ]]] b. [ RP boys of the boys ] = [ NP boys] - [ DP the [ NP boys] ] The NP boys denotes a set of plural boys (a join semi-lattice, containing all the possible pluralities that can be assembled with some number of boys in the domain, including the singularities but excluding the empty plurality). The definite the boys denotes the maximal element of the semilattice (i.e. the largest plurality of boys in the domain). The result of the subtraction is the set of all pluralities minus the largest one. This result is then fed to a numeral. Syntactically, this means that in partitives Num selects an RP which has acquired nominal features from the NP in its specifier, and filters away any plurality which is incompatible with the cardinal it contains. For example, with three individual boys a, b and c in the domain, the derivation for two boys 12

would be: (39) a. boy = a, b, c b. [ NP boys] = a,b,c, a,b, b,c, a,c, a, b, c c. [ DP the [ NP boys]] = Max( [ NP boys] ) = a,b,c the extract the largest plurality as in Sharvy 1980 d. [ RP [ NP boys] [ R ¼ [ DP the boys]]] = [ NP boys] - [ DP the boys] = a,b,c, a,b, b,c, a,c, a, b, c - a,b,c = a,b, b,c, a,c, a, b, c partitive denotation e. [ NumP 2 [ RP boys of the boys]] = a,b, b,c, a,c f. [ DP D ¼ 2 boys of the boys] Spell-out This account derives various features of partitives. The subtractive semantics obtains proper partitivity, i.e. the observation that a partitive is always smaller than the definite it embeds (see (40a)). Since the denotation of RP does not contain a supremum, the account derives the impossibility of definites to combine with partitives (40b); since proper names, though definite, do not have an NP which can move to [Spec,RP], partitives based on conjoined proper names are excluded (40c). For other aspects and details of the proposal, see Zamparelli (1998). (40) a.??two / One of my two parents. b.??the 3 of the boys came. c.??two of John, Jack and Mary The question is whether this syntax and semantics can be applied to our kind-denoting definite. Three ingredients are necessary. One is a phonetically null version of the operator defined above. The second is the assumption that the application of is restricted, perhaps even a lastresort operation, used when necessary to prevent a crash in the derivation. The third ingredient is the idea that in a Romance kind-denoting definite such as i cani the dogs, the NP cani denotes a singleton containing an individual kind-sort, here marked dogs. This way, the meaning of the definite article can be kept constant across normal and kind-denoting definites; the difference is localized in the argument the definite article receives. (41) [ DP i [ NP cani]] = Max( dogs ) = dogs What happens if we plug this DP in the Residue Phrase? Copying the internal NP to [Spec,RP] and performing the subtraction, we get the empty plurality: (42) [ RP [ NP cani] [ R ¼ di [ DP i [ NP cani]]]] = dogs - [ DP i cani] = dogs - dogs = But is not an acceptable restrictor for D. In order to give this construction a well-formed meaning is forced to apply to cani as the NP is merged in [Spec,RP]. As a result of this application [Spec,RP] will denote not a kind, but the set of individuals which belong to it. (43) [ RP [ NP cani][r ¼ di [ DP i [ NP cani]]]] = cani - [ DP i cani] = x dog ¼ (x) - dog = x dog ¼ (x) 13

Since the set containing the dog-kind is not a subset of the set of its instances, the result of performing the subtraction is once again the set of all individual dogs. This is the result we want. In more informal terms, the intuition behind this account is that the di we see in dei/desnominals is the same D of partitives. When we attempt to apply partitive di to something which has no parts a kind-entity we force a shift to the closest thing to a partitive meaning kinds can have: the set of instances which make them up. 10 In what follows I will adopt this system; for consistency, I will continue to mark the Residue Phrase as a PP. 5.1 The problem of singulars One remaining problem for a kind-based analysis of dei/des-nominals is why the application of di to other kind-denoting nominals doesn t give the same meaning or distribution as dei-nominals. For instance, the construction we have seen in (36) above cannot appear in argument position. Contrast (44a and b): (44) a. Fido è [della razza di cui parlavamo]. Fido is [of the breed we talked about] b. *A proposito di animali, ho visto [della razza di cui parlavamo] speaking of animals, I saw [of the breed we talked about] it should mean:...i saw one of the breed we talked about Moreover di cannot combine with another construction which appears to kind-denoting, the singular definite generic we see in brackets in (45a). Plugging this meaning under di, we might expect to get the impossible argumental and predicative cases in (45a,c). (45) a. [Il leone] è il re della foresta. [the lion] is the king of the forest b. *[Del leone] era il re della foresta. [of the lion] was the king of the forest c. *Simba era [del leone] Simba was [of the lion] My proposal is that the source for the ill-formedness of (44)b and (45) is the fact that they are both based on singular count Ns (razza breed and leone lion ). 11 In other terms, the di+il complex raising to D with singular semantic number is not capable of licensing the empty Num and D positions: (46) A di+def-art with singular count features cannot license D/Num *[ DP de+la Ò [ NumP t [ PP t [ DP t razza]]] 10 Another option is of course to derive the set of subkinds for that kind and then perform the subtraction, a possibility which does occur, but which I set aside in this paper. 11 In addition, (45) might also be excluded by independently documented problems in applying to a singular definite generic (see Krifka et al. 1995 and Dayal 2000). 14

The rationale behind this proposal (from Heycock and Zamparelli 2000) is that one role of NumP is that of turning the set denoted by its complement into an unsaturated property, a suitable restrictor for the determiner. To serve this semantic role, an empty Num head must be activated, either by checking plural/mass features or via selection by a lexical determiner in D. The dei complex raises to D to check features but it doesn t select the NumP projection in any sense, so the presence of a plural/mass feature on dei as it transits through Num becomes crucial. Pulling together the discussion so far, we arrive at the following formulation for the DP dei ragazzi: (47) Syntactic Structure for Kind-based dei/des-nominals: a. [ PP di [ DP¾ i [ NP ragazzi]]]]]] Base b. [ PP [ NP ragazzi] [ P ¼ di+i [ DP¾ i ragazzi]]]]] Copying ragazzi and the article c. [ DP½ dei [ NumP dei [ PP [ NP ragazzi] [ P ¼ dei [ DP¾ i ragazzi]]]]] Merging NumP, DP, moving dei and erasing intermediate copies for Spell-Out In another departure from Chierchia s account, I will assume here that dei and des are agreeing forms of the proposition di/de, and the agreement is triggered by movement of the definite determiner onto P (see Napoli and Nevis 1987). This formulation avoids the problem of why the movement of D to P doesn t seem to obey the mirror principle (which should give *i+di, on a par with the German pronoun+preposition cases davor, lit. that-from ). 6 On the absence of singular bare partitives As we have seen, the idea that dei/des-nominals are derived from kinds requires an independent assumption, stated in (46), to exclude various singular count cases. Does the original partitive account fare better in this respect? I don t think so, since even in this approach, singular count nouns must be excluded in some special way; this is because there is a type of partitive in Romance, the so-called proportional partitives which can contain a singular definite DP (see (48a)). Yet, this construction cannot be the basis for a bare partitive such as (48b): (48) a. Metà del ragazzo era sott acqua. half of the boy was under water b. *Del ragazzo era sott acqua. of the boy was under water How the absence of (48b) should be approached in a non-kind-based partitive account depends on what (48b) should mean. If (48b) were to mean part of the boy, it could be excluded by stipulating that the part-of relation Chierchia attributes to the empty N of bare partitives cannot be mereological part. But if (48b) means a boy, with the part-of boiling down to identity (as in Storto s example (14)b above, see footnote 2), (48b) would presumably be excluded by a notion of economy of structure: since Italian has a singular indefinite article, it doesn t use the structurally more complex singular bare partitive. 15

Economy is a delicate matter, and there is no room for a full discussion here. However, it is instructive to consider for a moment which determiners should to be compared for the purpose of economy. In Italian, all the forms listed in (49) are possible; evidently, alcuni some doesn t compete with dei and with the bare plural here. (49) Ho visto alcuni ragazzi / dei ragazzi / ragazzi I have seen some boys / of the boys / boys This is as it should be: the meaning of the three objects is close, but not identical, since these forms have distinct scopal and binding properties: dei differs from alcuni in that it can be bound by generic operators (see (20)), and it differs from the bare plural in the possibility to take wide scope. Whatever their origin, these meaning differences must be what allows these forms to coexist side by side even in an economy-ruled language. Now consider French. In French, as Storto points out, plural/mass des-nominals have obligatorily narrow scope, like Italian or English bare plurals. The singular indefinite article, on the other hand, has the same range of meanings as its Italian or English counterpart; it can have wide or narrow scope. The question is what rules out (50) in French: (50) *Je I voudrais peindre du palais. would like to paint of the palace where du palais should mean a palace, with narrow scope. It cannot be economy alone since the closest correspondent, the singular indefinite article, does not have the same range of meanings: unlike the bare partitive, it can have wide scope. Thus something extra needs to be said, most likely something which appeals to the singular count number features, just like the proposal in (46). 7 Conclusions In this article I have proposed a compositional derivation for the plural/mass indefinite determiners in Italian and French. The main idea is that these forms are derived by applying a partitive operator to a definite which denotes an individual kind. The result of this operation is the set of individuals which instanciate the kind. Combined with the idea that the des/dei-complex raises to the Num or D position of the upper DP, this analysis offers an account for the distribution and scope of dei/des-nominals in Italian and French, and for various features of the definite which appears to be embedded in these nominals. References Barker, C. (1998). Partitives, double genitive and anti-uniqueness. Natural Language and Linguistic Theory 16, 679 717. 16

Carlson, G. N. (1977). Reference to Kinds in English. Ph. D. thesis, University of Massachusetts at Amherst. Casalegno, P. (1987). Sulla logica dei plurali. Teoria (2), 125 143. Chierchia, G. (1998a). Partitives, reference to kinds and semantic variation. In A. Lawson (Ed.), Proceedings of Semantics And Linguistic Theory, Volume VII, Ithaca, NY. Cornell University: CLC Publications. Chierchia, G. (1998b). Reference to kinds across languages. Natural Language Semantics 6, 339 405. Cinque, G. (1994). Partial N-movement in the Romance DP. In G. Cinque et al. (Eds.), Paths Towards Universal Grammar, pp. 85 110. Washington, D.C.: Georgetown University Press. Contreras, H. (1986). Spanish bare NPs and the ECP. In I. Bordelois, H. Contreras, and K. Zagona (Eds.), Generative Studies in Spanish Syntax. Dordrecht: Foris. Dayal, V. (2000). Number marking and (in)definiteness in kind terms. Ms., Rutgers University. Delfitto, D. (1993). A propos du statut lexical de l article partitif en francais: Quelques hypothèses sur l interaction entre morphologie et forme logique. In A. e. a. Hulk (Ed.), Du lexique à la morphologie: Du cˆoté de chez Zwaan. Amsterdam: Rodopi. Heycock, C. and R. Zamparelli (2000). Coordinated bare definites. Ms., University of Edinburgh, Università di Bergamo. Hoeksema, J. (Ed.) (1996). Partitives: Studies on the syntax and semantics of partitive and related constructions. Berlin: Mouton de Gruyter. Kakegawa, T. (2000). Noun phrase word order and definiteness in Japanese. In Proceedings of the 19th West Coast Conference of Formal Linguistics., pp. 246 259. Krifka, M. et al. (1995). Genericity: An introduction. Chapter 1, pp. 1 124. Chicago, IL: University of Chicago Press. Longobardi, G. (1994). Proper names and the theory of N-movement in syntax and logical form. Linguistic Inquiry 25, 609 665. Napoli, D. J. and J. Nevis (1987). Inflected prepositions in italian. Phonology Yearbook 4, 211 228. Renzi, L. (1995). L articolo. In L. Renzi, G. Salvi, and A. Cardinaletti (Eds.), Grande grammatica Italiana di Consultazione, Volume III. Bologna: Il Mulino. Roy, I. (2001). Weak des/du-nps in French and judgement forms. Ms., USC. Selkirk, E. (1977). Some remarks on noun phrase structure. In P. Culicover, T. Wasow, and A. Akmajan (Eds.), Formal Syntax, pp. 285 316. London: Academic Press. Sharvy, R. (1980). A more general theory of definite descriptions. The Philosophical Review 89.4, 607 624. 17

Storto, G. (2001). On the status of the partitive determiner in italian. Ms., UCLA (in http://semanticsarchive.net). Zamparelli, R. (1995). Layers in the Determiner Phrase. Ph. D. thesis, University of Rochester. (Published by Garland, 2000). Zamparelli, R. (1998). A theory of Kinds, Partitives and OF/Z Possessives. In A. Alexiadou and C. Wilder (Eds.), Possessors, Predicates and Movement in the Determiner Phrase, Volume 22 of Linguistics Today, pp. 259 301. Amsterdam: John Benjamins. Zamparelli, R. (2001). Definite and bare kind-denoting noun phrases. ms., to appear in: Natural Language and Romance Linguistics II, Proceedings of Going Romance 2000. Eds. Frank DRijkoningen, Claire Beyssade, Paola Monachesi and Reineke Bok-Bennema, John Benjamins Publishers, Amsterdam/Philadelphia. Zucchi, A. (1995). The ingredients of definiteness and the Definiteness Effect. Natural Language Semantics 3, 33 78. 18