Noun phrases, such as this photo, can often appear both with modifiers, such as the adjective old, and with arguments, typically PPs such as of the queen. Most of the literature on the syntax of DPs has adopted the hypothesis that adjectives are base-generated higher than nominal arguments, just as adverbs are merged higher than arguments in the clause. This squib looks at the syntax of the Italian DP, focusing on the distribution of modifier PPs and argument PPs, and offers novel evidence that, contrary to the received view, nominal modifiers such as adjectives are in fact base-generated lower than nominal arguments.
Much of the literature on the syntax of DPs (e.g., Giorgi and Longobardi 1991, Alexiadou, Haegeman, and Stavrou 2007, Longobardi and Silvestri 2013) assumes that nominal arguments are merged lower than modifiers such as adjectives. The aim of this squib will be to argue that this assumption is fundamentally incorrect.
A typical implementation of the “received” approach is exemplified in (1).
[DP D [NumP Num [FP AP F* [α POSS [SUBJ [OBJ . . .
N . . . ]]]]]]
(adapted from Longobardi 2001:597)
An intuitive advantage of (1) is that it establishes a clear parallelism between the verbal and nominal extended projections, insofar as the lowest domain of both CPs and DPs is assumed to be the θ-domain (labeled α in (1)), where arguments are introduced. This is dominated by an “inflectional” domain where adverbs (in CP) or adjectives (in DP) are introduced.
The surface word order in English, where argument PPs occur to the right of the noun and adjectives occur to its left, can be simply derived by head movement of the noun to a position just above α. In Italian, on the other hand, the noun head moves higher, above FP, thus deriving the fact that (most) adjectives are postnominal in Italian, like PPs (see Longobardi 2001, Longobardi and Silvestri 2013). It is worth noticing that the approach in (1) goes hand in hand with the idea that linear order is fundamentally isomorphic with structural height, so that constituents that occur farther right (e.g., PPs) are systematically structurally lower. This will play an important part in my criticism of in section 2.
In this squib, I will present some Italian data that challenge the approach in (1), and I will argue, in line with Adger (2013), that arguments in the DP are merged higher than modifiers, including (most) attributive adjectives. I will limit my attention to nonderived nominals: if the arguments of deverbal nouns are introduced by verbal functional structure (Borer 2013b), they will not be true nominal arguments and may thus be expected to exhibit different syntactic behavior.
Some terminological clarifications are in order. The distinction between arguments and modifiers, primarily semantic in nature, will be at the heart of my discussion. The separation of adnominal constituents into these two classes has a long history. For example, Munn (1994), focusing on English “Saxon genitives,” distinguishes between regular possessives, as in John’s shoes, and modificational possessives, as in men’s shoes (‘shoes for men’). Kolliakou (1999), focusing on French de-phrases, distinguishes between individual-denoting expressions and property-denoting expressions. Finally, Borer (2013a), focusing primarily on Hebrew construct states, contrasts I(ndividual)-genitives and I-constructs with M(odificational)-genitives and M-constructs.
In this squib, I will use the term argument for expressions denoting themes, agents/authors, and possessors. These are referential or quantificational expressions that can bind anaphors and antecede pronouns (see the cited literature for details and other diagnostics). As argued in Longobardi 1994, 2008, semantic argumenthood requires the projection of a full DP, rather than a bare NP. This, however, can in turn be embedded within a PP or AP layer,1 resulting in “prepositional” or “adjectival” arguments, respectively. On the other hand, I will use the term modifier specifically for those intersective or subsective expressions that denote a property (e.g., size and color adjectives, as in small red book) or restrict the denotation of the modified noun to one of its subclasses (classifying adjectives, as in nuclear physics and dark comedy; see Rutkowski and Progovac 2005). I will not use this term for any other adnominal constituents, such as demonstratives, cardinals, ordinals, or high scope-taking adjectives (see footnote 3). Thus defined, both arguments and modifiers crosscut the categorial distinction between APs and PPs.
To highlight my semantically oriented perspective, I will focus only on arguments and modifiers that uniformly surface as PPs headed by the preposition di ‘of’ in Italian. This will allow me to preempt any attempt to explain away my data by appealing to different linearization properties of attributive adjectives and PPs (cf. Belk and Neeleman 2015); my claim is ultimately not about a categorial distinction (AP vs. PP), but about a semantic one, and about the way it is mapped onto the syntax of nominal structures.
2 Evidence from Word Order
The first piece of evidence for my claim comes from the neutral order of postnominal arguments and modifiers in Italian, represented schematically in (2).
una statua biancaAP di marmoPP(MOD) di CanovaPP(ARG)
a statue white of marble of Canova
‘a white marble statue by Canova’
The subscripted numbers in domain I of the template refer to the hierarchy of modifiers in (4) and indicate that hierarchically lower modifiers systematically occur closer to the head noun in linear terms.
SUBJECTIVE COMMENT > SIZE > WEIGHT > AGE > COLOR > ORIGIN > MATERIAL > CLASSIFYING
(adapted from Scott 2002:114)
There are many approaches in the literature to deriving the hierarchy in (4). Concerning APs, a common cartographic approach is to analyze modifier adjectives as specifiers of dedicated functional heads that merge rigidly in the order in (4) (see Cinque 1994, Scott 2002, Laenzlinger 2005). A radically opposed alternative is to attribute the ordering of adjectives purely to extrasyntactic cognitive preferences for adjectives introducing more “apparent” or “objective” properties to occur closer to the noun (see Sproat and Shih 1987, Scontras, Degen, and Goodman 2017). Finally, there are approaches that do derive the order in (4) in the syntax, but emphasize the way in which the semantics of adjectives directly determines their syntactic position (e.g., Bouchard 2002, Svenonius 2008). In Svenonius 2008, for example, adjectives adjoin to independently motivated projections in the DP that gradually build up the denotation of the whole nominal extended projection, and the semantic contribution of each adjective determines which projections it can adjoin to felicitously. Whatever the correct account of (4) might be, what matters for present purposes is that the hierarchy is real and appears to restrict possible orders for modifiers regardless of their categorial status: both modifier APs (5) and modifier PPs (6) abide by it (see Scott 2002 for a similar point).
uno strumento musicaleCLASS indianoORIG anticoAGE
an instrument musical Indian ancient
‘an ancient Indian musical instrument’
una tazza di ceramicaMATER di colore rossoCOLOR di due
a mug of ceramic of color red of two kiliWEIGHT
‘a red ceramic mug that weighs 2 kilos’
The template in (2) also shows that arguments are rigidly ordered, following the hierarchy in (7). Exactly as in the case of modifiers, arguments lower on the hierarchy appear closer to the noun, as (8) exemplifies.
POSSESSOR > AUTHOR/AGENT > THEME
(based on Longobardi and Silvestri 2013:89)
una statua di BaccoTHEME di MichelangeloAUTH di
a statue of Bacchus of Michelangelo of
‘a statue of Bacchus by Michelangelo that belongs to Alice’
The hierarchies in (4) and (7) have been argued to be first-Merged (see Scott 2002, Longobardi and Silvestri 2013); consequently, it appears that, within each of the three domains in (2), the lower an argument or modifier is merged, the closer it will appear to the head noun linearly.
The fact that postnominal arguments and modifiers occur in the mirror image of their first-Merge orders should immediately cast some doubt on the approach in (1). Mirror-image effects call for an analysis whereby constituents appearing farther right are higher than constituents appearing farther left, as in an account based on roll-up movement (see Shlonsky 2004, Cinque 2010). If this is the case, the fact that argument PPs occur farther right than modifier PPs (as well as modifier APs), as shown in (2), indicates that they are structurally higher, contrary to (1). This is the first evidence that modification takes place lower than argument introduction in the DP, evidence that will be corroborated below.
It should be noted that the order of postnominal constituents in (2) can be permuted. However, this does not happen freely: deviations from the neutral word order are usually the result of focusing a constituent to the right, as in (9).
il ritratto di MariaTHEME PRINCIPALE/*principale
the portrait of Maria main
‘the MAIN portrait of Maria’
Alternatively, constituents that can be used predicatively may appear farther right than expected if they are introduced as part of a reduced relative clause. (For an extensive discussion of this strategy and the ways to diagnose it, see Cinque 2010.) Needless to say, all the judgments reported in this squib refer to the neutral word order under a wide focus reading, unless stated otherwise.
3 Diagnosing Constituency: Evidence from Scope, Ellipsis, and Coordination
In section 2, I established (2) as the neutral order of postnominal constituents in the Italian DP and I suggested that the most natural analysis, based on roll-up or an equivalent mechanism for linearization,2 places arguments in a higher structural position than modifiers. In this section, I will look for confirmation that the more peripheral constituents in (2) are also structurally higher than those that are closer to the head noun and, in particular, that argument PPs are higher than modifier PPs, as represented schematically in (10).
(10) [DP D0 [[[N0] PPMOD] PPARG]]
A first piece of evidence comes from the interpretation of some scope-taking adjectives.3 (11) shows that prenominal presunto ‘alleged’ can scope either over the noun and both its argument and its modifier (11c) or over the noun and the modifier to the exclusion of the argument (11b). What is crucial is that presunto cannot scope over the noun and the argument to the exclusion of the modifier PP (11d). This suggests that a constituent structure as in (10) is on the right track. Note that nothing in my discussion predicts the deviance of (11a), which might be attributed to constraints against merging the scope-taking adjective presunto too low.
Ellipsis data also confirm the structure in (10). NP-ellipsis can target the noun only (12a) or the noun and the modifier PP to the exclusion of the argument PP (12b). Crucially, however, it cannot target the noun and its argument to the exclusion of the modifier PP (12c).
Yet another test that appears to confirm the structure in (10) comes from DP-internal coordination, as shown in (13). Following Longobardi (1994) and Bouchard (2002), such cases involve a single DP within which some nominal subconstituents are coordinated: syntactically, this is indicated by the fact that there is a unique determiner; semantically, the structure has a unique referent, to which multiple descriptions apply. As expected, coordination in such cases can apply to nouns alone (13a) or to nouns and their modifiers to the exclusion of their arguments, which are introduced higher up, above the entire coordination (13b). Interestingly, as (13c) shows, coordination cannot instead apply to nouns and their arguments to the exclusion of their modifiers.
4 Evidence against an “NP-Shell” Account
The constituency relations in (10) seem well-supported. However, it is important to rule out the possibility that, alongside the structure in (10), schematized in (14), the order in (2) might also be derived through a kind of “NP-shell” derivation, as proposed in Belk and Neeleman 2015, schematically represented in (15).4
The mirror-image effects discussed in section 2 already cast some doubt on the viability of (15), which cannot easily derive them. Moreover, if (15) were a possible underlying structure for (2), we would expect to be able to merge a scope-taking adjective such as finto/falso ‘fake’ between PPMOD and PPARG, where it could scope rightward over PPARG (and the trace of N0) only. However, this is not possible: whether finto/falso is linearized to the left (16b) or to the right (16c) of the argument PP, it systematically takes scope leftward.
il gioiello finto/falso di ambraMATER di mia
the jewel fake of amber of my
‘my grandma’s fake amber jewel’
Only reading: of my grandma, of amber > fake > jewel
il gioiello di ambraMATER finto/falso di mia
the jewel of amber fake of my
‘my grandma’s fake amber jewel’
Only reading: of my grandma > fake > jewel, of amber
il gioiello di ambraMATER di mia nonnaPOSS
the jewel of amber of my grandma
‘the fake amber jewel of my grandma’
Only reading: fake > jewel, of amber, of my grandma
This pattern, whereby postnominal finto/falso always scopes leftward, is exactly what is expected under the constituency in (14) if finto/ falso is a rightward modifier, as demonstrated in (17).5 On the other hand, under the constituency in (15), we would expect finto/falso to be able to scope rightward as indicated in (18), contrary to fact.
The above test can also be used to diagnose constituency with multiple modifiers or multiple arguments, as shown in (19). The structure that can be deduced from this is represented schematically in (20), an expanded version of (10). The subscripts LOW and HIGH are merely notational shorthands to indicate a PP’s relative height in the hierarchies (4) and (7) with respect to other PPs on the same hierarchy, however one may wish to derive such hierarchies (as discussed above).
un gioiello di ambraMATER finto/falso di colore
a jewel of amber fake of color
‘a fake amber jewel, which is red’
of color red > fake > jewel, of amber
una foto di un fulmineTHEME finta/falsa di Man
a photo of a thunderbolt fake of Man
‘a fake picture of a thunderbolt by Man Ray’
of Man Ray > fake > picture, of a thunderbolt
(20) [DP D0 [[[[[N0] PPMOD(LOW)] PPMOD(HIGH)] PPARG(LOW)] PPARG(HIGH)]]
un gioiello di giadaMATER di colore verdeCOLOR e . . .
a jewel of jade of color green and
‘a green jade jewel and . . .
. . . uno
gioiello di giadaMATERdi colore biancoCOLOR
a jewel of jade of color white
. . . a white jade jewel’
. . . *uno
gioiello di colore verdeCOLORdi
a jewel of color green of
una foto de-lla montagnaTHEME di mia madreAUTH
a photo of-the mountain of my mother
e . . .
‘a picture of the mountain that my mother took and . . .
. . . una
foto de-lla montagnaTHEMEdi mia
a photo of-the mountain of my
. . . a picture of the mountain that my sister took’
. . . *una
foto di mia madreAUTHde-l boscoTHEME
a photo of my mother of-the woods
In the foregoing discussion, I have explored the order of postnominal argument and modifier PPs in (2) and argued that it should be represented syntactically as in (14) and (20), the rightward position of the PPs being derived through roll-up movement. If the PPs discussed above are all first-merged in their surface positions, nothing needs to be added in defense of the claim that modification takes place lower than argument introduction in the DP (as in Adger 2013).
5 Evidence from “Truncated” Structures
An alternative possibility, however, is that the surface order in (2) is derived through movement operations such that only the surface position of arguments is higher than that of modifiers. This is in fact what Cinque (2010) proposes, suggesting specifically that arguments are merged lower than modifier adjectives but end up surfacing higher as a result of a series of movement operations, following the approach to of-phrases in Kayne 2000a,b, 2005.
An interesting piece of evidence for placing not only the surface position but also the Merge site of nominal arguments higher than the Merge site of modifiers comes from cases of “truncation,” where only an NP rather than a full DP is projected. The pseudopartitive structure in (23a) is one such case (see Alexiadou, Haegeman, and Stavrou 2007, Rutkowski 2007). The complement of di ‘of’ in (23a), an NP, can include both modifier APs and modifier PPs. Crucially, however, it cannot include arguments, regardless of whether they are definite or indefinite, or whether they are PPs or “possessive adjectives,” as (23b) demonstrates.
un bicchiere di [NP vino sicilianoORIG / de-lla
a glass of wine Sicilian of-the
‘a glass of Sicilian wine’
*un bicchiere di [NP suoPOSS vino / vino di
a glass of her wine wine of
MariaPOSS / vino di un vicinoPOSS]
Maria wine of a neighbor
An argument can only be introduced in a full partitive structure (24), where the complement of di is a full DP, as indicated by the presence of the article (i)l ‘the’.
un bicchiere de-[DP -l suoPOSS vino / -l vino di
a glass of the her wine the wine of
MariaPOSS / -l vino di un vicinoPOSS]
Maria the wine of a neighbor
‘a glass of her/Maria’s/a neighbor’s wine.’
Another case where less structure than a full DP is required is the complement of some subtypes of modifier PPs headed by di, such as material PPs (25) and classifying PPs (26). Only modifier APs or PPs can appear within such complements (see (25a), (26a)). Nominal arguments can only be introduced in different structures, such as (27), where the preposition su ‘on’ takes a full DP complement, as the presence of the article indicates. Again, this is the case regardless of the categorial status (AP vs. PP) or the (in)definiteness of the arguments.
una tavola di [NP legno neroCOLOR / di colore
a table of wood black of color
‘a table (made) of black wood’
*una tavola di [NP suoPOSS legno / legno di
a table of her wood wood of
MariaPOSS / legno di un vicinoPOSS]
Maria wood of a neighbor
una lezione di [NP musica giapponeseORIG / de-l
a lesson of music Japanese of-the
‘a lesson on Japanese music’
*una lezione di [NP suaAUTH musica / musica di
a lesson of her music music of
MozartAUTH / musica di un compositore
Mozart music of a composer
una lezione su-[DP -lla suaAUTH musica / -lla musica
a lesson on the her music the music
di MozartAUTH / -lla musica di un compositore
of Mozart the music of a composer
‘a lesson about her/Mozart’s/an unknown composer’s
A final case where only a sublayer of the DP is projected is the complement of tipo di ‘kind of’, which Zamparelli (2000:99–116) argues to be a KIP (Kind Phrase), one of the lowest layers of the noun phrase. In what follows, I will stick to the label NP for this small sublayer of the DP. Again, the same pattern emerges: modifier APs and PPs are grammatical (28a); arguments are not (28b).6
Questo tipo di [NP teoria chomskianaCLASS / teoria
this kind of theory Chomskyan theory
di stampo chomskianoCLASS] è sorpassato.
of ilk Chomskyan is outdated
‘This type of Chomskyan theory is outdated.’
*Questo tipo di [NP teoria di ChomskyAUTH] è
this kind of theory of Chomsky is
The data presented in this section suggest that nominal arguments are merged into the noun phrase higher than the layer where modifiers are introduced and licensed. The latter is the only layer to be projected in the complement of the preposition di in (23), (25)–(26), and (28), making argument introduction impossible. It might be tempting to attribute the ungrammaticality of arguments in (23b), (25b), (26b), and (28b) to a mere failure to license them: we may conjecture that these “truncated” structures lack the relevant case-assigning functional heads (which would be higher in the DP), leaving any argument caseless. However, this cannot be right: case assignment seems in principle available even in such structures, as demonstrated by the grammaticality of the modifier PPs della Sicilia in (23a) and del Giappone in (26a), which clearly contain a full DP in need of case, just as the ungrammatical argument PPs di Maria in (23b) and di Mozart in (26b) do. The natural conclusion is that the ungrammaticality of (23b), (25b), (26b), and (28b) is due to a failure to introduce the arguments in the first place.
The purpose of this squib has been to argue, on the basis of an Italian case study, that nominal arguments are first-merged higher in the DP than nominal modifiers, contrary to the “received view” in the literature.
In sections 2–4, I have analyzed the neutral order of postnominal arguments and modifiers and applied a series of diagnostics to demonstrate that arguments surface higher than modifiers. In particular, I looked at the interpretive possibilities of the scope-taking adjectives presunto ‘alleged’ and finto ‘fake’, NP-ellipsis, and DP-internal coordination. In section 5, I analyzed a series of “truncated” structures, where only a sublayer of the DP is projected and where only modifiers seem to be acceptable. I took this as evidence that arguments not only surface higher than modifiers, but are also merged higher.
Throughout the squib, I focused on adnominal constituents uniformly realized as PPs headed by di, differentiated into those with the semantics of arguments and those with the semantics of modifiers. This is a semantic distinction that crosscuts surface syntactic categories such as A and P (see footnote 1); I chose to concentrate on PPs only in order to abstract away from syntactic phenomena that depend on the adjective/preposition distinction rather than the modifier/argument distinction. One such phenomenon is the fact that adjectives seem to always appear closer to the noun than PPs. For example, modifier PPs, despite having the same semantics and ordering restrictions as modifier APs, systematically occur to the right of the latter.
un vaso cineseORIG anticoAGE di porcellanaMATER
a vase Chinese ancient of porcelain
di colore rossoCOLOR
of color red
‘an ancient red Chinese porcelain vase’
This phenomenon calls for an explanation that goes beyond what was discussed in this squib, perhaps by appealing to the prosodic differences between adjectives and PPs. I leave this as a matter for further investigation, as my discussion has centered around a different distinction, a semantic one between modifiers and arguments.
A second issue that deserves further investigation is the behavior of deverbal nouns. If their arguments are merged as part of a verbal structure that is embedded within the nominal structure, we might expect those arguments to be lower than the noun’s adjectives and the arguments of nonderived nouns. I predict that the diagnostics discussed above, applied to deverbal nouns, will indeed show this. See Borer 2013b (esp. section 3.2) for a discussion of the differences between the arguments of deverbal and nondeverbal nouns along the lines sketched here.
My purpose in this squib has been to show that modification in the DP takes place systematically lower than argument introduction. I have not attempted to derive this fact here, but the discussion above suggests that a better understanding of the interaction between the semantics and the syntax might be crucial to shed light on this issue.
1 Italian “possessive adjectives” (e.g., suo ‘his/her’) are nominal arguments that are ultimately realized as A(P)s. They have the same semantics, referentiality, and binding possibilities as full PP arguments (see Longobardi and Silvestri 2013), indicating the presence of a D(P), yet they have the morphology of adjectives and undergo gender and number concord with the head noun.
2 For the purposes of this squib, I will keep appealing to the notion of roll-up movement, although without any theoretical commitment to it. Nothing in my argumentation hinges specifically on roll-up being the correct mechanism for deriving mirror-image effects and rightward modification (see, e.g., Abels and Neeleman 2009 for an alternative).
3 The scope-taking adjectives considered here (e.g., altro ‘other’, finto ‘fake’, presunto ‘alleged’) have a very different distribution from the modifier adjectives that (4) refers to (e.g., alto ‘tall’, rosso ‘red’, inglese ‘English’). I assume that the former can merge quite high in the DP, and in different positions depending on the intended scopal reading. The main claim of this squib is not intended to apply to them.
4 The simplified structures in (14) and (15) are adapted from Belk and Neeleman 2015. I have avoided labeling nonterminal nodes in the nominal extended projection in order to abstract away, for the moment, from the absolute height at which modifiers and arguments are introduced. Only their relative height should matter in the present discussion.
5 The dashed lines indicate possible structural positions for finto/falso (only the relevant ones are represented here, for reasons of space). The boxed constituents, on the other hand, indicate the possible scopal domains for finto/ falso, covarying with the adjective’s height.
6 Note that the classifying adjective chomskiana is a modifier, rather than an argument. First, it has a broader range of interpretations than arguments proper, and it cannot always be replaced with an argument PP. Second, its denotation cannot provide the linguistic context with a referent to antecede a pronoun: ?*That Chomskyi an theory made himi famous (and its Italian equivalent), uttered out of the blue, is not acceptable (see Arsenijević et al. 2014).