What’s hidden below definiteness and genitive: on indefinite partitive articles in Romance

Abstract

In French, Italian, and other Romance languages indefinite nominal phrases can be introduced by what appears to be the conflation of a genitive preposition and a definite article, the so-called “indefinite partitive articles” (e.g., Fr. Je cuisine de la soupe depuis deux jours. ‘I’ve been cooking soup for two days’). This is rather unexpected, since these nominal phrases are neither definite nor in a syntactic position in which we expect to find a genitive preposition. This led part of the literature to consider them as built by lexical items synchronically distinct from the genitive preposition/definite article but homophonous with them. This contribution shows how a constituent-based approach to the lexicon-syntax interface as nanosyntax, paired with a specific take on the sequence of syntactic functions, can capture their apparently conflicting distribution without stipulating multiple homophonous lexical items. The key factor in this proposal is a revised analysis of the Romance lexical item (LI) for (i) definite articles – linked to a constituent containing not only features of definiteness but also lower indefinite features and higher nominative/accusative case features – and (ii) the genitive preposition DE – linked to a constituent containing not only genitive features but also lower nominative/accusative features. Holding these LIs crosslinguistically stable, the variation attested in this domain is modeled as depending on the amount of functional structure lexicalized by the nominal root in the different languages.

1. Introduction

Indefinites in Romance languages have raised substantial interest in the last few decades, especially regarding their crosslinguistic variation (Cardinaletti and Giusti 2016; Delfitto and Schroter 1991; Espinal and Cyrino 2021, 2022; Stark 2008a, 2008b, 2016) and their morphosyntax (Chierchia 1997; Zamparelli 2008). As for the first point – the crosslinguistic variation – it has been observed that the same indefinite context is marked differently in Italian and French. While in Italian (1) the noun minestra ‘soup’ is not introduced by any overt functional element (i.e., the noun is bare, B[are] N[oun] henceforth), in French (2) we find the overt marker de la, literally ‘of the’.

(1).

Cucino	minestra	da	due	giorni. (It.)
cook.1sg	soup	from	two	days
‘I have been cooking soup for two days.’

(2).

Je	cuisine	de	la	soupe	depuis	deux	jours. (Fr.)
sbj.1sg	cook.1sg	of	the.f.sg	soup	from	two	days
‘I have been cooking soup for two days.’

In this case, Italian is akin to English, as the translation shows. However, Italian possesses a form that is parallel to French de la ‘of.the’ and that, like its French counterpart in (2), may also be used in indefinite contexts (3):

(3).

Ho	cucinato	della	minestra. (It.)
have.1sg	cooked	of.the.f.sg	soup
‘I cooked some soup.’

In this respect, the main points of interest are how to model this crosslinguistic variation and what it tells us regarding the structure of indefinites.

A closer look at the indefinite markers in (2) and (3) leads us to the second topic: their morphosyntactic structure. These elements are morphologically complex, being composed of the genitive preposition de ‘of’ – i.e., the form used to introduce genitive nominal phrases (see de ‘of’ in [4]) – and the definite article – i.e., the form used to introduce nominal phrases referring to maximal or unique entities (see le and la in [4]).

(4).

Le	jouet	de	la	fille. (Fr.)
the.m.sg	toy	of	the.f.sg	girl
‘The toy belonging to the girl.’

This makes (2)–(3) superficially identical to partitive structures like (5) introducing a subset of a definite set (partitives henceforth, see Jackendoff 1968; Selkirk 1977, among others), leading part of the literature to label them “indefinite partitive articles” (indefinite PAs henceforth) (Pinzin and Poletto 2022a, 2022b; Stark 2008a, 2008b, 2016, among others).

(5).

J’	ai	mangé	une	partie	de	la	soupe	qu’	ils	m’
sbj.1sg	have.1sg	eaten	a	part	of	the.f.sg	soup	that	sbj.3pl	dat.1sg
ont	donnée. (Fr.)
have.3pl	given
‘I ate a part of the soup they gave me.’

In this respect, the focus is on the relationship between their morphological and semantic composition (see Section 3 for a literature review on the topic). First, a genitive preposition is not expected to introduce a direct argument (subject or object). Second, the presence of a definite determiner seems to clash with the general semantics of these constituents, which does not include any reference to a maximal entity in the discourse, as apparent from (2)–(3).

In this contribution, I put forward a novel proposal regarding both the syntactic structure of indefinite PAs and what regulates their crosslinguistic distribution. The core of the proposal is that both the lexical item (LI henceforth) for de ‘of’ and for the definite article correspond to complex syntactic structures in the lexicon (in line with nanosyntax; Starke 2009, 2014). More precisely, I suggest that (i) the LI for de ‘of’ contains not only the syntactic feature for genitive case, but also the syntactic features for nominative and accusative case (assuming Caha’s 2009, 2013, 2019 Case Containment Hypothesis) and (ii) the LI for the definite article contains, in addition to the features allowing it to refer to maximal entities in the discourse, the features for lexicalizing different layers of indefiniteness. I will show how this proposal, in combination with the nanosyntactic lexicalization algorithm (see Caha 2019; Starke 2018), allows these LIs to surface in the different syntactic contexts highlighted in (1)–(5). As a second step, I show how the crosslinguistic variation across Romance languages in this domain can be modeled as depending on the amount of structure directly lexicalized by the LI that corresponds to the noun in each language, in line with other recent proposals in the literature which bind the variation in the lexicalization patterns to differences in the number of functional heads lexicalized by roots (among others, see Caha et al. 2019; Vanden Wyngaerd et al. 2020 for adjectival morphology in Slavic; Caha et al. 2023 for adjectival/verbal zero derivation in English and Czech; Janků 2022 for nominal declensions in Czech; Bertocci and Pinzin 2020 for verbal thematic vowels in Latin; Cortiula 2023 for verbal morphology in Friulian).

Such an analysis has positive consequences on at least two fronts. On the one hand, it reconciles an approach adopting single LIs for de ‘of’ and definite articles with much of the data showing that indefinite PAs (2)–(3), prepositional phrases (4), and partitives (5) have different syntactic distributions and behave differently when we submit them to the same range of tests (see Section 3): allowing different syntactic structures to be lexicalized by the same LIs solves this apparent paradox. On the other hand, it enables us to model the attested crosslinguistic variation across different Romance languages purely based on the shape of the LIs without referring to syntactic parameters or other sources of variation (i.e., the syntax of indefinites is kept constant in all languages analyzed), in line with the Borer-Chomsky Conjecture (Borer 1984).

The paper is organized as follows. In Section 2, I present the data and the empirical boundaries of the present investigation. In Section 3, I go through the main approaches and proposals regarding the analysis of indefinite PAs in Romance languages. Section 4 is dedicated to an overview of nanosyntax, the framework adopted for the present contribution. Section 5 presents the core of the argument, where I offer a novel approach to indefinite PAs and their crosslinguistic variation, especially focusing on their competition with BNs and definites. In Section 6, I highlight a shortcoming of the proposed approach related to the divide between structural and oblique cases, suggesting a possible solution. Finally, Section 7 draws some conclusions.

2. Crosslinguistic variation

In this section, I provide an overview of the crosslinguistic variation addressed in this paper, specifying the limits of the current analysis. From a syntactic-semantic point of view, I focus on the expression of two kinds of indefinites. The first type is represented by indefinites referring to an open-ended quantity, whether mass or count, ranging over any possible subset of the referent set denoted by the property-noun. More concretely, I look at the externalization of sentences like I’ve been [building houses/cooking rice] since I was born, examining how houses/rice is lexicalized in different languages. As a second type of indefinites, I focus on those referring to indefinite quantities (numbers or amounts) whose consistency is determined but unspecified (I baked [some] bread). I define these two indefinites in a more precise way in Section 5.2, but the general idea is that in the second context we have a boundary defining the indefinite quantity we are dealing with, while in the first case this boundary is absent (see Pinzin and Poletto 2022a, 2022b for other data; Tortora 2008 for an analysis of [un]boundedness in the spatial domain). Henceforth, I refer to the first type as unbounded indefinites, while I refer to the second type as bounded indefinites. I further constrain the analysis to positive environments (see Baldi and Savoia 2022; Garzonio and Poletto 2020; Savoia and Baldi 2023 for an overview of the negative patterns in the Italo-Romance varieties) and to the Romance varieties in which one of or both the highlighted indefinite types present either DE + def (= indefinite PAs) or the bare preposition DE, as defined in the following paragraph.

I use a set of descriptive labels to identify potentially similar objects in different languages, which consistently appear in the relevant indefinite contexts. With the label DE I refer to the prepositions derived from Latin de ‘of’ used in genitive contexts, as marking the possessor in possessive contexts. I extend this descriptive label to every occurrence of this form, irrespective of other properties.

(6).

La	casa	di	Piero. (It.)
the.f.sg	house	of	Piero
‘Piero’s house.’

With the label def I refer to definite determiners, the set of markers which appear when the DP refers to a maximal entity (see Section 4). Again, I extend this label to every occurrence of these forms, irrespective of other properties.

(7).

Ho	incontrato	la	vicina	di	Ugo. (It.)
have.1sg	met	the.f.sg	neighbor	of	Ugo
‘I met Ugo’s neighbor.’

With the label DE + def I refer to markers consisting of the conflation of a DE element with a def element (e.g., Fr. de la [f.sg], du [m.sg]; Emil. adla [f.sg], dal [m.sg] etc.), as, for example, when the possessor is referentially definite in a possessive relationship.

(8).

La	casa	della	vicina. (It.)
the.f.sg	house	of.the.f.sg	neighbor
‘The house of the neighbor.’

Finally, with the label bare noun (BN) I refer to the cases in which the noun is not introduced by any overt marker, including DE, def, or their combination.

(9).

Mangio	fragole	da	una	vita. (It.)
eat.1sg	strawberries	from	a	life
‘I’ve been eating strawberries since forever.’

Romance languages vary with respect to the way in which they externalize bounded and unbounded indefinites by means of DE, def, and BN. Let us first look at unbounded indefinites (I’ve been building houses for years). A first group of languages – exemplified in (10) by Italian – shows BNs in these cases.

(10).

Costruisco	case	da	anni. (It.)
build.1sg	houses	from	years
‘I have been building houses for years.’

This does not hold for every Romance language. A second group of languages, among which French and Emilian, shows DE + def, see (11)–(12).

(11).

Je	construis	des	maisons	depuis	30	ans. (Fr.)
sbj.1sg	build.1sg	of.the.pl	houses	from	30	years
‘I’ve been building houses for 30 years.’

(12).

I	è	an	ke	kostruisi	adli	kà,	ma	a	n	ù
sbj.3pl	is	years	that	build.1sg	of.the.f.sg	houses,	but	sbj.1sg	not	have.1sg
mai	vist	atse	bröti. (Emil., Dosolo)
never	seen	so	ugly
‘I’ve been building houses for years, but I’ve never seen such ugly ones.’
(DiFuPaRo 405)

Finally, there are varieties where we only find DE. An example is the Franco-Provençal variety of Stnic, see də mɛnta ‘of mint’ in (13).

(13).

De	fɔʀijẽ	n	æn	kwiʎɛ	də	mɛnta	œ pœ	ˈĩ	sən′aː. (Fr.Pr., Stnic)
of	spring	sbj.1pl	have.1pl	collected	of	mint	for	one	week
‘In spring, we collected mint for a week.’
(DiFuPaRo 261)

If we consider bounded indefinites (I baked [some] bread) we have a different distribution. In this case, Italian behaves like French and Emilian, in that they all present DE + def, see (14)–(16):

(14).

Ho	cotto	del	pane. (It.)
have.1sg	cooked	of.the.m.sg	bread
‘I baked some bread.’

(15).

J’	ai	cuit	du	pain. (Fr.)
sbj.1sg	have.1sg	cooked	of.the.m.sg	bread
‘I baked some bread.’

(16).

U	kot	dal	pan. (Emil., Dosolo)
have.1sg	cooked	of.the.m.sg	bread
‘I baked some bread.’
(DiFuPaRo 438)

The Franco-Provençal variety of Stnic still behaves differently, presenting only DE in these environments too, see də tsu/kar′ɔt ‘of cabbage/carrot’ in (17).

(17).

Dze	dʒəndɔ	də	tsu	e	də	kar′ɔt. (Fr.Pr., Stnic)
there	add	of	cabbage	and	of	carrot
‘I add some cabbage and carrots.’
(DiFuPaRo 47)

All in all, we can see two main patterns for unbounded and bounded indefinites: Italian overtly differentiates between them; French, Emilian, and Franco-Provençal do not. Additionally, Franco-Provençal differs from French and Emilian in that it only shows DE, with no def. This is represented in Table 1.

Table 1:

Lexicalization of unbounded and bounded indefinites in the Romance languages under analysis.

	Unbounded indefinites	Bounded indefinites
Italian	BN	DE + def
French, Emilian (Dosolo)	DE + def	DE + def
Franco-Provençal (Stnic)	DE	DE

This typology exhausts the possibilities when looking at the Romance languages that present either DE + def or DE in either bounded or unbounded indefinites.

3. How many lexical items do we need?

The literature on indefinite PAs revolves around two main issues: (i) how to account for the crosslinguistic variation we observe in their distribution and (ii) how to account for the co-occurrence of what looks like a genitive preposition (DE) and what looks like a definite determiner (def) in these indefinite markers. The second issue can be further elaborated in the following way: how many LIs do we need? Can we unify the presence of DE and def in indefinite PAs with their tokens as used in other contexts? Cardinaletti and Giusti (2016, 2018, Ihsane (2008), Stark (2016), and Storto (2003) provide a negative answer to the question. In short, they propose that we are dealing with complex indefinite markers that do not actually contain any LI connected to the definite article or the genitive preposition. Storto (2003) proposes an analysis of indefinite PAs in terms of “lexical indefinite determiners” (Storto 2003: 330). In the same vein, for Stark (2016) and Ihsane (2008) DE in indefinite PAs is the lexicalization of a low classifying head within the functional sequence of the indefinite DP (ind[ividuation]° in Stark 2016; de° in Ihsane 2008), while def is a higher (agreement) marker into which DE incorporates. Finally, adopting the same general idea, Cardinaletti and Giusti (2016, 2018 argue that DE and def are lexicalizations of an indefinite DP structure, with def – an agreement marker – in the D° and DE in its specifier. Espinal and Cyrino (2021, 2022 differentiate themselves from these analyses in that they assume that there is no distinction between the instance of def in indefinite PAs and def as generally used for referring to definite entities. In other words, they do without assuming the presence of two different but homophonous def LIs for the definite article. In their analysis, indefiniteness is achieved by means of the DE element, defined as an abstract operator shifting definites into property-type expressions. To put it differently, they propose that (these kinds of) indefinites are derived from definites by canceling the definiteness restriction via the DE operator. While they unify all instances of def, they still maintain that DE in indefinite PAs is not synchronically related to the other instances of DE (genitive PPs/partitives), thereby assuming two homophonous DE LIs. In favor of these approaches assuming different LIs there is a set of syntactic arguments showing how indefinite PAs and genitive PPs/partitives have different behaviors and distributions. The first observation, already presented before, is that indefinite PAs have the distribution of direct arguments (subject or object) and not of PPs. A second related fact regards PP extraction (Espinal and Cyrino 2022; Ihsane 2008, 2013; Milner 1978). More specifically, indefinite PAs do not block PP extraction (18), while partitives and PPs block it (19).

(18).

C’	est	de	nos	poules	que	nous	mangeons	souvent	des	oeufs. (Fr.)
it	is	of	our	chickens	that	sbj.1pl	eat.1pl	often	of.the.pl	eggs
‘It is of our chickens that we often eat eggs.’
(Carlier 2007: 20)

(19).

*C’	est	de	Zola	que	j’	ai	lu	deux	des	livres. (Fr.)
it	is	of	Zola	that	sbj.1sg	have.1sg	read	two	of.the.pl	books
(Milner 1978: 71 ff.)

While the idea of separate homophonous LIs takes care of these cases in a straightforward fashion, it requires proposing a large-scale homophony “spreading” to different languages, even beyond the Romance continuum. This has been supported in terms of general shared diachronic processes. Carlier (2007) and Carlier and Lamiroy (2014) elaborate a proposal regarding the grammaticalization path of indefinite PAs in French, while Schurr (2021) proposes a comparative diachronic analysis within the Romance continuum (see Seržant 2022 for a wider comparative analysis beyond Romance languages).

In contrast to these approaches, a second set of proposals takes the different instances of DE – and not only def (Espinal and Cyrino 2021, 2022) – to be synchronically linked to a single LI in the lexicon, albeit with different degrees of “unity”. Some authors focus only on partitives versus indefinite PAs, unifying them and leaving aside the occurrences of DE as a genitive preposition. Chierchia (1997) and Zamparelli (2008), for example, propose an analysis of Italian indefinite PAs that implies this kind of partial unity. The gist of these proposals is that DE performs the same function in both partitives and indefinite PAs. More specifically, for Chierchia DE performs no semantic function, so that it has no meaning in either case, while in both cases def performs the same function it generally does in the language. This in turn means that, in his approach, there is no semantic difference between partitives and indefinite PAs. On the other hand, for Zamparelli, DE lexicalizes the same semantic operator, labeled R for “residue”. This operator returns the denotation of its specifier minus the denotation of the complement. The difference between partitives and indefinite PAs is then related to the denotation of the complement subject to the R operation: as a complement of R, partitives have a “regular” definite referring expression, while indefinite PAs have a kind-denoting DP (introduced by definite articles in this language; see footnote 1). Both Zamparelli and Chierchia do not explicitly account for the syntactic differences between indefinite PAs and partitives shown in (18)–(19) and for the occurrence of DE in genitive PPs (see [4], [6]–[8]). Manzini (2019), finally, proposes a generalized unity hypothesis. The core idea is that the LI for DE is always an element carrying relator content, defined as imposing a part/whole relationship (⊆). Its syntactic label is that of a preposition (P). When this element merges with a complement DP (e.g., [_P DE] + [_DP la viande]), the newly created constituent {DE, DP} is either labeled as a PP following DE ([_PP [_PDE] [_DP la viande]]) or as a DP following la viande ([_DP [_P DE] [_DP la viande]]). The choice between the two is context-dependent. Adopting this approach, the syntactic problem related to the extraction facts highlighted in (18)–(19) disappears. If the DE part/whole relator can be merged with a complement DP and the new syntactic object can receive the same label as the complement DP itself, then it can be selected by a verb, both in object and subject position. Along the same lines, it could be claimed that labeling the whole constituent PP blocks the extraction of a PP merged within it.

In short, we see a continuum of proposals ranging from the ones advocating for the necessity of having different homophonous LIs for both def and DE in the lexicon (Cardinaletti and Giusti 2016, 2018; Ihsane 2008; Stark 2016; Storto 2003; plus the diachronic analyses in Carlier 2007; Carlier and Lamiroy 2014; Schurr 2021) to the proposals assuming that both the def and the DE we find in indefinite PAs are the same elements we generally see in the language in other contexts (Manzini 2019). As a middle ground, Espinal and Cyrino (2021, 2022, Chierchia (1997), and Zamparelli (2008) support different degrees of “unification”.

The crosslinguistic variation within Romance languages is only modeled by a subset of these proposals: Cardinaletti and Giusti (2016, 2018, Espinal and Cyrino (2021, 2022, Stark (2016). While Stark (2016) draws a connection between the expression of plurality on the noun and the distribution of BNs and PAs in Romance languages (see also Delfitto and Schroter 1991), the other proposals relate the crosslinguistic variation to different language-specific lexicalization choices of the same structure. Crucially, no account assuming partial or total unity for the occurrences of DE (Chierchia 1997; Manzini 2019; Zamparelli 2008) addresses the issue of the crosslinguistic variation observed in Section 2. In what follows, I put forward a proposal which unifies the occurrence of DE and def in both unbounded and bounded indefinites with their other occurrences under a single LI for each, while at the same time accounting for the observed crosslinguistic variation.

Before going into the analysis, a few remarks on the adopted framework are in order.

4. Nanosyntax, an overview

Nanosyntax (Starke 2009, 2014) is a realizational model for the lexicon-syntax interface based on constituency that assumes that syntax is the only combinatorial mechanism within the set of modules producing linguistic expressions. The label “realizational” is shared with other approaches such as distributed morphology (Halle and Marantz 1993) and means that LIs are seen as realizing (= lexicalizing) syntactic-semantic units, which are merged previously and independently of them. Syntactic-semantic units are conceived as a set of content-bearing features (see [20]) combined by the speaker to output complex meanings. The combinatorial mechanism is assumed to be restricted to binary merge. This merge operation, being binary, operates by taking a single feature and merging it either with another feature or with the result of a previous merge operation. The result of a series of merging operations is therefore a nested set of features (21), referred to in syntactic terms as constituent or phrase.

(20).

{x, y, z …}

(21).

The order of merge of the different syntactic-semantic features is guided by the final meaning the speaker wants to express. Each merge operation modifies the meaning of the constituent built by the previous merge operation. A set of constituents in a specific order (and, therefore, with a specific meaning) makes up a functional sequence (fseq). For example, Caha (2009, 2013 advocates for a specific fseq for case systems, in which different cases correspond to different constituents in a containment relationship. What we generally label nominative corresponds to a constituent containing the feature {f1} ([nomP]), accusative corresponds to a constituent containing [nomP] and {f2} ([accP]), genitive corresponds to a constituent containing [accP] and {f3} ([genP]) and so on.

(22).

This means that a constituent containing only {f3} will not be interpreted as genitive. It is the composition of three features that is interpreted as such. In this case, the different fseqs for each case are different from one another because of the presence/absence of a feature on the top of the derivation. However, when we enlarge the picture to larger segments of the structure, it is easy to realize that fseqs frequently diverge because of a feature in the middle and not because of a feature on the top. In other words, different fseqs are generally opposed to each other because of gaps: one of the two fseqs has an additional feature in the middle of the sequence with respect to the other. Such gapped structure becomes visible when one adopts a layered structure for the (in)definiteness features, in line with Ihsane (2008, 2013 and Zamparelli (1998, 2000, among others. To see why, let us further propose that a definite derivation builds on an indefinite one by adding a feature (for a similar analysis see Zamparelli 2000). This additional feature can be thought of as imposing a maximality restriction {max} on the denotation of the noun (Fintel et al. 2014; Link 1983; Sharvy 1980; Wągiel 2021; Zamparelli 1998, 2000; see Section 5 for more details).

(23).

As commonly and independently assumed (see Lamontagne and Travis 1987; Picallo 1991, among others), the case fseq closes-off the nominal domain and is merged on top of the (in)definiteness one. In this situation, a “definite nominative” fseq – in (24) – will be different from an “indefinite nominative” one – in (25) – because of an additional feature in the middle.

(24).

(25).

Note that the label of the top constituent (“nomP”) is the same in both cases. This is a shortcut, adopted for expository reasons: the two “nomPs” constituents are different: one contains {max}, while the other does not.

As far as syntax per se is concerned, without taking into consideration the relationship with the lexicon, nothing more needs to be added: the syntactic component takes as input content-bearing syntactic-semantic features and gives as output nested sets of these features derived by a binary merge operation. In standard nanosyntactic practice, no crosslinguistic variation is attributed to this part of the linguistic computation. Languages are maximally homogeneous with respect to the available set of syntactic-semantic features and their ordering, so that a linguistic expression that has the same interpretation in two different languages will contain the same set of features in the same order, i.e., fseqs are universal (in line with cartographic approaches, see Cinque 1999; Rizzi 1997). All the crosslinguistic variation should be attributed to the language-specific lexicon involved in the externalization procedure, still to be outlined.

The externalization procedure takes care of lexicalizing the nested sets of syntactic-semantic features derived by syntax by putting them into communication with the lexicon of the language, i.e., a list of LIs. LIs are conceived as tripartite entities (26a) containing an externalization form (for spoken languages this is generally equal to a phonological representation), syntactic-semantic information and conceptual content, holding the information that is not encoded by syntactic-semantic features. The syntactic-semantic information is stored in the same form in which they appear in syntax: nested sets of features, i.e., constituents (“the (morpho)syntactic lexicon contains nothing but well-formed (morpho)syntactic expressions” [Starke 2014: 1]). Since the conceptual content is not relevant in the following discussion, I simplify LIs as in (26b) and refer to them by means of the externalization form, such as a in (26b).

(26).

The lexicalization (= externalization) procedure is cyclic and is triggered every time a feature is merged with the fseq at a given point in the derivation. Each time a feature is merged, the lexicalization procedure searches the lexicon for a LI matching the constituent derived by the syntactic computation. Each feature must be lexicalized for the derivation to be successful. A LI is considered an adequate match for a given constituent if it contains that constituent, but it can contain other features as well. This is the reasoning behind what is generally known as the Superset Principle in nanosyntax: “A lexically stored tree matches a syntactic node iff the lexically stored tree contains the syntactic node” (Starke 2009: 3). Let us pretend our toy language has the LI a in (26b) in its lexicon and that syntax derives [f1P], as in (27). In this situation, the LI a matches the syntactic derivation and lexicalizes it, as highlighted.

(27).

The syntactic component can then derive another constituent by merging {f2}, as in (28). The same LI a containing {f1} provides a suitable option in this case too, so that the lexicalization procedure is satisfied.

(28).

By these principles, the single LI a can lexicalize more than one fseq, giving rise to non-accidental syncretism or coexpression (i.e., more functions, one entry).

For now, we have only seen cases in which a single LI is adequate for lexicalization. It could be the case, however, that more LIs match the constituent derived by syntax. Let us enrich our lexicon with the LI b, as in (29). Given the same syntax as in (28), both a and b are adequate lexicalizations; they “compete”. The resolution of this competition comes from the application of the Elsewhere Principle (Kiparsky 1973, among others): use the more specific LI. Since b applies to a subset of the contexts in which a applies, b is more specific and wins the competition.

(29).

This principle is relabeled in nanosyntactic terms as Minimize junk (“At each cycle, if several LIs match the root node, the candidate with least unused nodes wins” [Starke 2009: 4]).

Another possibility is that no LI matches the constituent derived by syntax as it is. Here one of the core principles of nanosyntax comes into play: the cyclic lexicalization procedure must find an adequate match and each feature must receive an adequate lexicalization at each cycle, otherwise syntax does not proceed merging another one. To achieve this goal, the lexicalization procedure goes through a series of steps (i.e., an algorithm) to “fix” the syntactic derivation. After each step the lexicon is searched to see if it was successful. As soon as a step is successful (i.e., all features receive a lexicalization), the procedure is satisfied, and the syntax can go on merging another feature. The algorithm is assumed to be universal, so that no crosslinguistic difference is supposed to be encoded here. In this contribution I adopt the following version of the algorithm, argued for in Caha (2019) (but see Starke 2018).

(30).

Merge {f}:
a.	Lexicalize [fP].
b.	Lexicalization-driven movements:
	i.	Spec-to-spec: Move the Spec of the complement of {f}. Retry (a).
	ii.	Complement-to-spec: Move the complement of {f}. Retry (a).
c.	Backtracking:
	i.	Go back one step in the derivation and try the next option for that cycle.
	ii.	Try to merge {f} in the complex left branch and lexicalize it.
d.	Complex left branch: spawn a new subderivation providing {f}, lexicalize it and merge it with the main derivation, projecting [fP].

Up until now we have only seen lexicalizations successful at step (a), so that the following steps of the algorithm were not triggered (see [27]–[29]). The steps in (b) involve lexicalization-driven movements, i.e., the movement of either the Spec (b.i) or the complement (b.ii) of the syntactically derived constituent. These “rescue” movements change the geometry of the tree and give lexicalization another opportunity to succeed. The step in (c.i) involves going back one lexicalization cycle and choosing the next-best step available, a way to undo a derivation that led to a dead-end. For a discussion of these mechanisms and a potential alternative to backtracking see Blix (2021). I will not discuss them in any more detail here since steps (a.)–(c.i) will not be relevant in the discussion. This is because no step until (c.ii) will ever be successful with the kind of LIs involved in the lexicalization of the nominal phrases described in Section 2 (DE, def, etc.). On a descriptive level, these LIs always linearly precede the lexical core of the fseq – the noun LI – so that they can be defined, following Starke (2018), as “pre” LIs or complex left branches. In the current understanding, the distinction between the LIs preceding (“pre”) and following (“post”) the lexical core of the fseq is driven by the complexity of the structure below the lowest non terminal node (foot henceforth) of the constituent with which they are stored in the lexicon: “post” markers have a unary foot (see [31a], where [f1P] dominates only {f1}), while “pre” markers/complex left branches have a binary foot (see [31b], where [f1P] dominates {f1} and x).

(31).

This follows from the mechanics of lexicalization-driven movements. If any kind of lexicalization-driven movement displacing parts of the fseq to its left has happened, the right constituent will necessarily have a unary foot. This is because lexicalization-driven movements vacate the position from which they start without leaving an interpretable trace. On the other hand, this means that a LI with a binary foot as (31b) will never match any right constituent derived by lexicalization-driven movements. Since the LIs currently composing our lexicon are only ever used as “pre” markers and therefore have a binary foot, steps (a)–(c.i) of the algorithm will never be successful. This allows us to skip these steps for the rest of the contribution. As for the two remaining steps, the idea is that in case a complex left branch is already present in the derivation, the algorithm will first try to expand it with the new feature (c.ii), then, if this does not lead to a successful lexicalization, a new complex left branch will be derived in an auxiliary workspace (d). In case a complex left branch is not already present in the derivation, step (c.ii) vacuously applies and a new complex left branch is derived in an auxiliary workspace (d). In the next sections I will adopt these mechanisms to derive the patterns observed in Section 2, starting with unbounded indefinites (see Table 1).

5. Genitive and definite markers lexicalizing indefinites

Let us recall the main issues posed by indefinite PAs. We have what appears to be a genitive marker (DE) appearing in a context in which we would expect no genitive feature (nominative/accusative arguments) and what appears to be a definite article (def) in a context in which we would expect no definiteness (indefinite nominal phrases). This is illustrated with Emilian adli kà ‘houses’ in (12) repeated below as (32).

(32).

I	è	an	ke	kostruisi	adli	kà,	ma	a	n	ù
sbj.3pl	is	years	that	build.1sg	of.the.f.sg	houses,	but	sbj.1sg	not	have.1sg
mai	vist	atse	bröti. (Emil., Dosolo)
never	seen	so	ugly
‘I’ve been building houses for years, but I’ve never seen such ugly ones.’
(DiFuPaRo 405)

In line with part of the literature (Carlier 2007; Ihsane 2008; Stark 2016), I argue that the structure in (32) does not contain genitive and definite features interpreted as direct structural case and indefinite, respectively. On the contrary, I contend that DE and def in (32) lexicalize neither genitive nor definiteness features. I follow the analysis of Caha (2009, 2013 for case features, in which the different cases correspond to different cumulative sets of features (see Section 4, example [22]). I adopt a similar containment hypothesis for the features regulating (in)definiteness: definite reference is built on the syntactic structure that yields indefinite reference by restricting the possible interpretations to a strictly maximal object in the context (Fintel et al. 2014; Link 1983; Sharvy 1980; Wągiel 2021; Zamparelli 1998, 2000). In this section I only deal with what I defined before as unbounded indefinites, that is indefinites of the kind found in sentences like I’ve been building houses/eating rice for years. The feature building the denotation of unbounded indefinites is labeled {indef}, which will serve as a basis for the derivation of bounded indefinites too (see Section 5.2). Below it, we have the predicative core of the nominal structure, labeled NP. The difference between the denotations of NP and indefP follows the intuition in Dobrovie-Sorin and Beyssade (2012), where it is claimed that mass and plural BNs, at least in Romance, are amount-referring expressions containing an existential quantifier (see Longobardi 2001 for a similar approach). This accounts for the fact that they can occur as arguments, in contrast to pure property-denoting expressions like adjectives. In (33a) we have an accusative indefinite fseq and in (33b) an accusative definite fseq.

(33).

The appearance of DE and def LIs as a lexicalization of (33a) is due to the fact that these LIs lexicalize complex constituents which not only contain the features for genitive/definiteness, but also the lower features for nominative/accusative ({f1}–{f2}) and indefiniteness ({indef}). By the Superset Principle, these LIs can lexicalize these lower features too. In (34) I present a first approximation of the LIs ad (DE), al (def), and the noun in (Mantovano) Emilian. Note that ad and al have a binary foot and therefore can only be used as complex left branches. I leave gender and number aside and present for def a single LI with the externalization form of the masculine singular, al.

(34).

By means of the Superset Principle, ad and al can lexicalize any of the constituents they contain: ad can lexicalize both [accP] and [nomP], while al can lexicalize [indefP]. Hence it is not surprising to find them introducing nominative/accusative unbounded indefinites. In this case, the question is why we do not see these LIs everywhere. How can we constrain ad and al to lexicalize the indefinite fseq in (33a)? Without further specifications, we would expect syncretism between definite (33a) and indefinite (33b) fseqs: each of them should be lexicalized as ad + al + noun. Let us see step-by-step how under the current assumptions (33a) and (33b) would come to be lexicalized by the same LIs. As a first step let us start from the NP – the predicative core – shared by the derivations of definite and indefinite fseqs alike. The syntactic component derives the NP and looks for a lexicalization, finding the noun LI in the lexicon of the language (for which, see [34]), as shown in (35).

(35).

When {indef} is merged, as shown in (36), the lexicon is searched again for an appropriate lexicalization.

(36).

The steps of the lexicalization algorithm in (30) from (a) to (c.i) fail since no LI with a unary foot is present in the lexicon. Step (c.ii) fails because there is no complex left branch already available. The only option is (d) – opening an auxiliary workspace and deriving a constituent closed by the feature to be lexicalized ({indef}) – and (37) is derived. (37) can be fully lexicalized by al in (34b) for the left constituent and by noun in (34c) for the right constituent.

(37).

At this point, the definite and indefinite derivations diverge. In the indefinite derivation in (38) case features are directly merged, starting with {f1} deriving [nomP].

(38).

As before, the steps from (a) to (c.i) fail because there is no LI with a unary foot in the lexicon. Step (c.ii) is then attempted and {f1} is merged on top of the complex left branch, as (39) shows.

(39).

In this case too, however, there is no possible lexicalization: no LI in the lexicon in (34) matches such a complex left branch. Step (d) is then attempted, leading to the creation of another complex left branch with {f1} as its highest feature, as in (40). Given the presence of ad in (34a), and assuming the Superset Principle, (40) can be fully lexicalized as that complex left branch is contained within (34a).

(40).

With a successful lexicalization, the next case feature {f2} can be merged and the derivation can proceed (I skip this part of the derivation for reasons of space; in this case, step [c.ii] is successful: the highest complex left branch is expanded with {f2} and ad lexicalizes it).

Let us see what happens with a definite derivation, starting from where we left it in (37). In this case, {max} is merged before case features, deriving (41).

(41).

The steps from (a) to (c.i) fail and (c.ii) is attempted, merging {def} on top of the complex left branch. Given the presence of al (34b) in the lexicon, the step is successful, yielding (42).

(42).

At this point, case features are merged, starting from {f1}. In our lexicon in (34), only ad contains {f1}. Nothing should therefore change from (40): only step (d) can be successful, yielding a new complex left branch lexicalized by ad, as in (43).

(43).

This is not adequate, however, given that in these languages a definite fseq is lexicalized as al + noun and not as ad + al + noun, as shown by al pütel ‘the boy’/li sigoli ‘the onions’ in (44).

(44).

Al	pütel	a	i	à	butadi	via	li	sigoli. (Emil., Dosolo)
the.m.sg	boy	sbj.3sg	obj.3pl	has	thrown	away	the.f.pl	onions
‘The boy threw away the onions.’
(DiFuPaRo 597)

The analysis makes incorrect predictions. As a first approach, one could postulate the existence of a phonologically null LI for nominative/accusative case, as in (45). This LI would outcompete ad in (34a) for the lexicalization of {f1}–{f2}, being more specific (by Minimize junk).

(45).

This solution would produce the expected Ø + al + noun lexicalization for the definite fseq in (33b), but it would also produce the same Ø + al + noun lexicalization for the indefinite fseq in (33a), leading to incorrect predictions again. The relevant fact to consider is that the presence of ad depends on the presence in the fseq of the feature {max}. That is, ad only appears when al does not lexicalize {max}, while ad is not necessary when al lexicalizes {max}. Constituent lexicalization gives us a straightforward explanation for this behavior. Let us propose that the LI al can lexicalize, on top of {max}, the nominative/accusative case features {f1}–{f2} too. We then substitute (34b) with (46) in the lexicon of this language.

(46).

In a definite derivation, where {max} is present, the LI al in (46) becomes a competitor for the lexicalization of {f1}–{f2} and outcompetes ad. This is because the lexicalization algorithm prefers expanding an already present complex left branch – step (c.ii) – to opening another auxiliary space and deriving a second complex left branch – step (d). The result can be seen in (47), where I show the full derivation up to the accusative feature {f2}.

(47).

On the contrary, the LI al in (46) would not be able to lexicalize {f1}–{f2} in an indefinite derivation. Recall that an LI can only lexicalize a constituent if it contains that constituent. If the derivation lacks {max} and merges {f1} directly with [indefP], the LI containing {max} in its lexically stored fseq in (46) is blocked from lexicalizing {f1}. Let us see this in more detail. The indefinite derivation gets to the stage in (38), repeated in (48).

(48).

By the logic of the lexicalization algorithm, all the steps are attempted. The steps up to (c.i) fail, as always in the cases at stake given the absence of LIs with a unary foot. Step (c.ii) is then attempted, deriving (49) (repeated from [39]).

(49).

Just as in (39), this step fails. There is no LI in the lexicon containing [nomP] on the left. Even the updated al in (46) fails, since [nomP] does not contain {max}, while (46) does. In other words, the gap (absence of {max}) blocks the lexicalization potential of al to [indefP], so that ad must come into play to lexicalize {f1}, being the only available competitor. The final syntactic configuration predicted for unbounded indefinites in Emilian and French is therefore the one already presented in (40), where the lower layers up to indefP are lexicalized by def, while the higher case layers are lexicalized by DE.

To summarize thus far, by virtue of two ingredients, (i) the constituent lexicalization algorithm and (ii) the LIs for def and DE in (46) and (34), we get an analysis for the main issue we started from: indefinite PAs contain definite and genitive LIs without any semantically interpreted definite/genitive feature being present. In doing so, the analysis allows for modeling the different behavior of DE + def in indefinite PAs versus genitive PPs/partitives observed in (18)–(19): indefinite PAs, unlike genitive PPs/partitives, do not contain any genitive syntactic-semantic feature {f3}. Binding the block on extraction to the presence of non-structural case features like genitive predicts the observed behavior, in line with Kayne (1975: 114–115), who suggests that the block stems from the presence of a PP node (i.e., genitive, in our terms) blocking the extraction of another PP embedded within it. If no PP node is present (i.e., no genitive, in our terms), extraction of an embedded PP is allowed (see also Cardinaletti and Giusti 2016: 65, who reach a similar conclusion assuming that PPs are strong islands for extraction).

In the next Section I model the crosslinguistic variation regarding the externalization of unbounded indefinites in the languages of the Italian and Franco-Provençal type (see Table 1).

5.1. Unbounded indefinites: crosslinguistic variation

The current analysis accounts for the fact that some languages, such as French and Emilian, lexicalize the fseq for unbounded indefinites in (33a) with a DE + def sequence, without there being any genitive ({f3}) or definiteness ({max}) feature in the syntactic derivation. The final syntactic configuration up to the nominative layer is in (50) (repeated from [40]).

(50).

However, in the same unbounded indefinite contexts in which Emilian and French have DE + def, languages such as Italian have BNs. See (9), repeated here as (51).

(51).

Mangio	fragole	da	una	vita. (It.)
eat.1sg	strawberries	from	a	life
‘I’ve been eating strawberries since forever.’

I propose to relate this crosslinguistic variation to the amount of structure that the nominal root can lexicalize in the different languages. This is in line with other proposals which bind the variation in the lexicalization patterns to differences in the number of functional heads lexicalized by the adjectival/nominal/verbal root (see Caha et al. 2019; Vanden Wyngaerd et al. 2020 for adjectival morphology in Slavic; Caha et al. 2023 for adjectival/verbal zero derivation in English and Czech; Janků 2022 for Czech nominal declensions; Bertocci and Pinzin 2020 for verbal thematic vowels in Latin; Cortiula 2023 for verbal morphology in friulian). The nominal root in French and Emilian is constrained to the lexicalization of the [NP] denoting a property, as in (34), so that for all the subsequent features additional LIs are needed. Italian, instead, has bigger nominal roots, which can also lexicalize {indef} and the case features {f1}–{f2}, in addition to the [NP] predicative core. This is shown in (52c). The LIs for def and DE in (52a) and (52b) are the same as for the other languages.

(52).

(52c) is a perfect match for the fseq in (33a) and, accordingly, fulfills step (a) of the lexicalization algorithm (direct lexicalization, with no movements) at each lexicalization cycle, outcompeting all other LIs. The result is shown in (53).

(53).

Note that (52c) can only be used if the syntactic derivation produces this exact constituent. In a definite derivation, {max} is merged with [indefP], yielding [defP]. Since (52c) does not contain [defP], it cannot lexicalize the derived constituent. In such a derivation, which I skip for reasons of space, the LI in (52a) must come into play, resulting in the syntactic configuration in (54).

(54).

By the same logic, we can derive the Franco-Provençal pattern, which for unbounded indefinites has only DE and no def (see [13]). As for Italian, we keep constant all LIs but the one for the noun, which in Fr.Pr. stops lexicalizing at [indefP] and is therefore smaller than Italian but bigger than French/Emilian, as in (55).

(55).

The resulting syntactic configuration is shown in (56), where noun and de lexicalize the whole structure.

(56).

By allowing the number of functional heads that nominal roots can lexicalize to vary between languages, we derive the different patterns observed for unbounded indefinites (the first column of Table 1). The small nominal roots in French and Emilian (see [34c]) need both def and DE for lexicalizing the indefiniteness layer and the case layers, respectively; the slightly bigger nominal roots of the Franco-Provençal varieties (see [55]) only need DE for lexicalizing the case layers; finally, Italian has the biggest roots (see [52c]) and, accordingly, does not need other LIs. The LIs for DE and def are kept constant across languages.

5.2. Bounded indefinites: adding a second layer

If we take into consideration bounded indefinites (I baked [some] bread), we have a different crosslinguistic distribution of the LIs (see Table 1). More specifically, while French, Emilian, and Franco-Provençal keep the output constant (DE + def, DE + def, and DE, respectively), Italian differentiates bounded indefinites from the unbounded ones analyzed in the previous section, employing DE + def instead of BNs ([57] = [14]).

(57).

Ho	cotto	del	pane. (It.)
have.1sg	baked	of.the.m.sg	bread
‘I baked (some) bread.’

I propose that on the syntactic-semantic level bounded indefinites are more complex and that the difference can be formally encoded as an additional feature on top of {indef}. To identify this feature, let us focus on the different semantics of indefinite BNs and PAs in Italian. As already hinted in Section 1, I claim that indefinite PAs, in Italian, introduce a boundary to the amount/number of the element(s) having the property denoted by the noun. Let us consider the following examples in which we have the consumption verb mangiare ‘to eat’ in two different structures, one with the clitic pronoun si ‘self’ (the same clitic used in reflexive constructions) and one without. As independently supported in the literature, in these contexts the clitic si is resultative, meaning that it derives an event in which the object is consumed in its entirety (see Folli and Harley 2005).

(58).

a.

Da	qualche	mese,	Gianni	mangia	(del)	gelato	ogni	giorno (It.)
from	some	month	Gianni	eats	of.the.m.sg	ice cream	every	day
‘It’s a few months that Gianni eats ice cream every day.’

b.

Da	qualche	mese,	Gianni	si	mangia	*(del)	gelato	ogni
from	some	month	Gianni	self	eats	of.the.m.sg	ice cream	every
giorno. (It.)
day
‘It’s a few months that Gianni eats ice cream every day.’

While the possibility of using indefinite PAs is unaffected by the presence or absence of resultative si, the same is not true for BNs. BNs are ungrammatical in the context in which resultative si is present, as (58b) shows. These facts can be interpreted as following from the fact that (58b) requires the object to be delimited, to have boundaries, so that it is possible to consume it in its entirety. Notice, moreover, that we are not dealing with a requirement on the specificity of the object, given that both contexts in (58) denote a series of daily events in which the subject eats some ice cream each time, which cannot be the same specific ice cream. The ungrammaticality of the BN in (58b) can then be related to the fact that BNs lack a boundary, which is an essential semantic component of these structures, thus restricting the lexicalization possibilities of the object. This also derives the incompatibility of Italian indefinite PAs with long-term habitual activity contexts like I’ve been building houses for years, where BNs are instead fully acceptable, as shown in (59a–b). In (59), the speaker is asserting that she has been doing the activity of building houses for years. The use of an indefinite PA in (59b) – which implies a boundary – makes it impossible to access this interpretation and instead forces an infelicitous one, in which the speaker asserts that she has been building an indefinite, bounded set of houses (specific or not) for years.

(59).

a.

Costruisco	case	da	anni. (It.)
build.1sg	houses	from	years
‘I have been building houses for years.’

b.

Costruisco	delle	case	da	anni. (It.)
build.1sg	of.the.f.pl	houses	for	years
#‘I have been building houses for years.’

Again, if we take indefinite PAs to be bounded, as opposed to BNs, we can explain this: the boundary imposes an interpretation of the object as an indefinite number of houses while the context requires an unbounded interpretation of the object as referring to any possible subset of the whole referent set of case ‘houses’.

Bearing these facts in mind, I label the feature differentiating these two types of indefinites as {boundary}, and the resulting phrase as [b(ounded)-indefP]. To reflect this formally, (60) presents the complete set of fseqs: unbounded indefinites are in (60a) and do not have the additional [b(ounded)-indefP] layer, while bounded indefinites and definites are in (60b) and (60c), respectively, and have the additional [b(ounded)-indefP] layer on top of [IndefP], hosting {boundary}. All fseqs are closed off by the case features, as has been argued for in Section 4. This highlights the different sizes of the gap: with respect to a definite fseq, unbounded indefinites lack two features in the middle ({boundary}, {max}), while bounded indefinites lack one feature ({max}).

(60).

​

The facts regarding Italian, then, can be captured by updating the lexicon in (52) to the one in (61), where the only difference is the additional presence of [b-indefP] in (61b) (cfr. [52b], where no [b-indefP] is present). (52a) and (52c) remain intact.

(61).

This lexicon, in conjunction with the lexicalization algorithm, derives the observed patterns. The noun LI in (61c) can lexicalize {indef} and the case features {f1}–{f2}, leading to the presence of BNs. This is only possible, however, when the fseq does not involve {boundary} or {max}, i.e., it is only possible with unbounded indefinites (60a). When {boundary} is merged and [b-indefP] is derived, we need to lexicalize it by means of the def LI il in (61b), which is the only one including the relevant feature. As shown before, then, if we are deriving a bounded indefinite (60b) – i.e., the fseq lacks {max} – the DE LI di in (61a) must be used to lexicalize {f1}–{f2}, yielding di + il + noun (62).

(62).

Note that the reason why the DE LI lexicalizes {f1}–{f2} in (62) instead of the def LI is the same as in (50) before (and [65] below): the relevant indefinite derivation contains a gap with respect to the definite derivation, i.e., it lacks {max}. As the def LI (61b) has {max} in between the indefiniteness layers and the case layers ({f1}–{f2}), it does not properly contain a constituent closed off by the case features but lacking {max}. By the Superset Principle, then, (61b) cannot lexicalize such constituent. If that is the case, di is inserted. By the same principles, if we are in a definite derivation – i.e., a derivation involving {max} – the LI il in (61b) can go on lexicalizing {f1}–{f2}, yielding il + noun.

As for the other varieties under analysis, we just need to update the LIs for accounting for the lexicalization of b-indefP. More specifically, for Emilian and French we must update the LI for def as we did for Italian (see [63b]), keeping the other LIs intact.

(63).

The noun LI is constrained to the [NP] level and, as in Italian, def can lexicalize {f1}–{f2} only if {max} is present. This yields ad + al + noun for both unbounded and bounded indefinites and def + noun for definites.

For Franco-Provençal, by contrast, we must update both the LI for def, as for the other varieties, and the noun LI, so that it can lexicalize {boundary} as in (64).

(64).

This yields de + noun for both the unbounded (60a) and the bounded (60b) indefinite fseq and def + noun for the definite fseq (60c). This follows from the fact that the noun LI can lexicalize every feature up to {boundary}, while {f1}–{f2} are always lexicalized by the DE LI as in (65), unless the feature {max} is merged.

(65).

Note that at this point we derived the crosslinguistic variation shown in Table 1. The observed patterns reduce to how much of the fseq the noun can lexicalize in the different languages, in line with other current proposals (Bertocci and Pinzin 2020; Caha et al. 2019, 2023; Cortiula 2023; Janků 2022; Vanden Wyngaerd et al. 2020, among others). The LIs for DE and def remain crosslinguistically constant.

6. Splitting structural and oblique cases

In the previous sections, I only dealt with instances in which bounded and unbounded indefinites have structural case, that is nominative/accusative. Let us consider what happens in Italian in contexts in which non-structural cases are involved. I will illustrate this using data from Italian, but, as far as I can tell, this extends to all languages analyzed in this paper. In general, Italian introduces datives with the preposition a ‘to’. Keeping the noun constant (the proper name Maria in this case), a alone marks dative, while di alone marks genitive. No sequence a + di is attested.

(66).

Ho	dato	la	palla	a	Maria. (It.)
have.1sg	given	the.f.sg	ball	to	Mary
‘I gave Mary the ball.’

(67).

La	palla	di	Maria. (It.)
the.f.sg	ball	of	Mary
‘Mary’s ball.’

Following, among others, Caha (2009, 2013, 2019, the set of case features following nominative/accusative directly builds on [accP], so that we have the fseq in (68).

(68).

Dative is built on the genitive by adding {f4} and {f5}. With this in place and within the adopted framework, the simplest analysis of the substitution pattern in (66)–(67) is to claim that the two markers do not co-occur because di (69a) is properly contained in a (69b).

(69).

The LI di, being more specific, wins the competition up to [genP], while a wins as soon as we merge {f4} and {f5} and derive [datP], overriding di. This should extend to the use of di as an accusative/nominative marker too. In other words, assuming the case fseq in (68) and the LIs in (69), the prediction is that a dative bounded indefinite fseq should be lexicalized in Italian, as a sequence a + def + noun (i.e., with a overriding di for the lexicalization of the case fseq). This is shown in (70).

(70).

The prediction is wrong, however. A and di co-occur exactly in the indefinite contexts we are looking at, where we find the sequence a + di.

(71).

Ho	dato	la	palla	a	delle	ragazze. (It.)
have.1sg	given	the.f.sg	ball	to	of.the.f.pl	girls
‘I gave the ball to some girls.’

This is not a peculiarity of a, since it happens with all other prepositions too: da/con/per delle ragazze ‘from/with/for of.the.f.pl (= some) girls’ (see Carlier 2007; Ihsane 2008 for similar remarks on French). More simply, di co-occurs with these other prepositions only in those contexts in which I claim that it lexicalizes a structural case, either nominative or accusative. There is then a split between how “structural” and “genitive” di behave. In principle, this issue can be dealt with either by modifying the LIs or the fseq. One approach could be to revive the homophony idea, claiming that we have two homophonous LIs di, one specified for nominative/accusative (72b) and the other for genitive (72a). The LI for a – (72c) – would then contain only (72a), “genitive” di.

(72).

This would superficially fix the problem, at the cost of giving up the idea I put forward in this contribution, that is, to adopt a unique lexical entry for DE (and def) in all its uses, proposed at the outset of the paper as an alternative to the homophony approach that makes use of at least two LIs for DE. Moreover, note that by (72a–b) we would have two homophonous LIs lexicalizing contiguous pieces of the fseq. To avoid this outcome, I propose modifying the fseq instead. The observed pattern splits “structural” di from “oblique” di. Let us then hypothesize that the case fseq, in these cases at least, is not contiguous as in (68) but split in two, with an additional feature – here labeled {z} – intervening between accusative and the rest of the case sequence, as shown in (73).

(73).

Let us further assume that the only LI that contains {z} in the lexicon is a phonologically null LI with a complex foot (i.e., a “pre”-marker).

(74).

In this scenario, the LI in (74) would block any interaction between the higher part of the case fseq and the lower part. It follows that there is no stage of the derivation in which the higher case features ({f3}, {f4}, etc.) are merged directly on top of the pre-marker di lexicalizing [accP]. As a result, a would not be able to overwrite “structural” di.

(75).

While {z} would shield an occurrence of “structural” di from being overwritten by a, the same would not happen if di is merged above it (i.e., it is an “oblique” di), since in that case the higher case features would be directly merged on top of the pre-marker di. Following this hypothesis, one could maintain a single di LI and explain the different patterns occurring when it lexicalizes structural and oblique cases.

An evident drawback of this proposal is that it requires stipulating an ad hoc null pre-marker in the lexicon, which would satisfy an unspecified feature {z}. However, the stipulation of a null pre-marker is not needed a priori. It could be possible to achieve the same effect by assuming that {z} is lexicalized by a feature-driven movement of a constituent from within its complement [accP]. This displaced constituent would block the interaction between the “structural” and the “oblique” layer as much as a phonologically null pre-element. Since there is still no full-fledged proposal incorporating feature-driven movement into the lexicalization algorithm in the literature (but see De Clercq 2020 for a potential approach) and developing one is beyond the scope of this paper, we set this issue aside for future work.

7. Conclusions

In this paper I put forward a novel proposal regarding the morphosyntactic structure of indefinite partitive articles in Romance languages and, from a broader perspective, why we find what otherwise appears to be a genitive preposition (DE) in nominative/accusative contexts and what otherwise appears to be a definite article (def) in indefinite contexts. In doing so, I maintain that the lexicon of these languages contains a single LI for each of DE and def. Note that, within the boundaries of the present data, this fact holds crosslinguistically too: the lexically stored tree for the LI DE is identical in each language analyzed, and the same is valid for def. The core of the proposal is that the DE and def LIs contain, respectively, nominative-accusative ({f1}–{f2}) and indefinite ({indef}-{boundary}) features, in addition to the features characterizing them as genitive ({f3}) and definite ({max}). This is exemplified in (76a–b) with the LIs for the Mantovano Emilian variety.

(76).

The fact that in French and Emilian the two types of indefinite fseqs analyzed – unbounded and bounded indefinites (Section 5.2) – come out as a sequence of DE + def while definite fseqs just as def is a consequence of constituent lexicalization. The LI def can lexicalize the {f1}–{f2} case features only if it also lexicalizes the definiteness feature {max}. This means that a definite fseq with structural case will surface as def alone, without any additional marker. If, on the other hand, {max} is not merged in the fseq (i.e., we have an indefinite fseq, unbounded or bounded), the LI DE wins the competition for the lexicalization of {f1}–{f2}, yielding the sequence DE + def. In the paper, I also highlighted how different Romance languages vary with respect to how they lexicalize unbounded and bounded indefinites (Section 2). I propose that this crosslinguistic variation can be accounted for in terms of the “size” of the nominal root, that is the number of functional features that it can lexicalize. French and Emilian have the “smallest” nominal roots, confined to lexicalizing the property [NP] (63c). This results in both bounded and unbounded indefinites being lexicalized as DE + def + noun. Some other languages, such as some Franco-Provençal varieties, have nominal roots growing to [b-indefP] (64). Such roots are therefore able to lexicalize both layers of indefiniteness but need the LI DE to lexicalize nominative/accusative case features, yielding DE + noun for both bounded and unbounded indefinites. Finally, languages such as Italian have nominal roots growing up to nominative/accusative case features but only incorporating the features for lexicalizing the layer for unbounded indefinites (61c). This results in a surface pattern in which we see just the noun (BN) as a lexicalization for unbounded indefinites and DE + def + noun as a lexicalization of bounded of indefinites.

Supplementary Material

Supplementary Material

Supplementary Material

This article contains supplementary material (https://doi.org/10.1515/ling-2022-0059).