Romance pronominal clitics as pure heads

Abstract

Romance clitics are currently accounted for as DP arguments moved to functional head positions or as functional heads (AccVoice, etc.) licensing pro-DPs in argument position. I take the view that clitics are first merged as heads, projecting independently motivated categories on the functional spine of the sentence (φP, ApplP). I argue that they can satisfy theta relations without need for a pro associate. From an empirical point of view, a pure head syntax for clitics is favoured in explaining the asymmetries between clitics and phrases, found in several syntactically relevant domains (order, agreement, case). I show how the hypothesis that clitics are functional heads derives the internal order of the clitic string, which does not necessarily match (or mirror) that of phrasal constituents. I also consider agreement asymmetries (perfect participle agreement) and case asymmetries (in relation to Differential Object Marking).

1. Outline

By the term clitic, I mean heads with a syntactic distribution that differentiates them both from inflectional affixes and from independent words, essentially the special clitics of Zwicky (1977). These elements are attested in many languages and language families (e.g. Indo-European, Semitic, Bantu, Pama-Nyungam) though they are sometimes called clitics, sometimes object markers. Romance pronominal clitics therefore exemplify a typologically significant phenomenon with universal implications.

There are two standard accounts of (Romance) clitics, namely the movement analysis of Kayne (1975, 1991), also adopted by Chomsky (1995), and the clitic-pro analysis of Sportiche (1996). Both accounts include a phrasal component, either because clitics are generated as DPs (movement theory), or because they licence pro-DPs. Schematically, the movement theory of clitics says that clitics are merged as DPs (or in the version of Chomsky 1995, as both maximal and minimal Ds) and then moved to functional head positions (adjoined to I in Kayne’s 1991 version) as in (1a). Sportiche’s theory has clitics first merged as functional heads (AccVoice etc.) and doubling phrasal pro’s in thematic position, as in (1b). ²

The purpose of this article is to argue that clitics can and should be treated as pure heads. In the present account, clitics are first merged as heads, cf. (2a), very much as proposed by Sportiche. However, they do not project specialized functional labels (e.g. AccVoice in (1b)), but rather their intrinsic categorial content, hence φP and ApplP, constructing part of the functional spine of the predicate in the process, cf. (2b) and (2c).

For instance, the accusative clitic is associated with a structure like (3), where φP is simply the node projected upon agreement of the internal argument with v* (Chomsky 2015).

The first question that arises with respect to the proposal in (2)–(3) is whether it is feasible. A potential problem is that in order to maintain (2)–(3), without any further additions, it is necessary to show that theta relations are well defined in structures like (3), in other words, that (4) also holds.

A different issue is whether (3) is advantageous. In very general terms, there is an element of simplicity in (3), which is represented precisely by the lack of devices (movement or pro) required to normalize the syntax of clitics to phrasal syntax. Empirically, any asymmetry observed between clitics, or a subset of them, and their putative phrasal counterparts is potentially an argument in favour of (3).

The article addresses the various points just raised, as follows. In Section 2, I provide an illustration of the asymmetries between clitics and phrasal arguments (including pronouns), focusing on order relations (depending on dominance, reflected by linearization). In Section 3, assuming that the asymmetries are best dealt with in the syntax, I address the question in (4). In Section 4, I argue that the internal orders of the clitic string reflect the order of independently postulated functional projections of the verb. Case and agreement asymmetries between clitic and phrasal syntax are reviewed in Section 5. Finally, clitics do not normally surface in vP, as in (3), but rather in IP – posing the question of the further derivation that structures like (3) must undergo. I will briefly mention possible answers in Section 6. ³

2. Asymmetries between clitics and phrasal arguments: previous accounts

Clitics are known to display a number of behaviours, including relative order, mutual exclusions, syncretisms, and suppletion, that distinguish them from phrasal pronouns and stand in the way of the conclusion that they are straightforward head versions of pronominal XPs. Let me begin by illustrating just one problem, namely order, which has a strong claim to involve the syntax. By order, I understand dominance, as possibly reflected by linear order via the linearization algorithm at EXT (Externalisation, Chomsky 2013).

Italian and French have the same order of direct object and indirect object DPs in phrasal syntax. Specifically, these two arguments are in a reciprocal c-command relation, since (unlike in English) a quantifier in either of them can bind a variable in the other, as in (5) for Italian and (6) for French. Though the linear order is normally DO > IO, as displayed, it can be inverted without changing the binding facts (see Folli & Harley 2006 on Italian; Boneh & Nash 2012 on French).

Now, accusative (Acc) clitics and dative (Dat) clitics are differently ordered with respect to one another in Italian and French, as in (7). We just saw that the order of these elements in phrasal syntax is identical and that there is no independent indication that movement parameters hold between the two languages – a point to which I will return. Therefore, clitic order does not reflect phrasal order.

While (5)–(6) versus (7) show that the same phrasal syntax can yield different clitic orders, the reverse is also true. Selected and unselected datives have different orders with respect to accusatives in both Italian and French phrasal syntax (Folli & Harley 2006; Boneh & Nash 2012). While in (5)–(6) selected datives (i.e. indirect objects) are in a mutual c-command relation with accusatives, the unselected datives (benefactives) in (8)–(9) are higher than direct objects, since quantifier-variable binding is possible only from benefactives to direct objects as in (8a) and (9a) and not the reverse.

Yet, when we turn to clitics, the same order found in (7) with indirect object clitics, is found in (10) with benefactives, in both Italian and French.

In short, the contrast between (5)–(6) and (8)–(9), taken together with the invariance of the corresponding clitic strings, shows that different phrasal syntaxes can yield the same clitic order. In fact, I don’t know of a single (Romance) language that splits dative clitics according to the selected/unselected distinction. Note that French does have two clitic slots for datives. However, it allocates them on the basis of Person, namely 3rd person dative following the accusative, and 1st/2nd person dative preceding accusative, as in (7b) and (11) respectively. Datives are never ordered according to the Person hierarchy in phrasal syntax (in Romance).

Recall then that in movement theories of clitics, clitics are phrasal arguments at first merge and reach their surface position via movement, as schematized in (1a). The syntactic derivation may involve head movement, i.e. D-movement, based on Chomsky’s (1995) proposal that a clitic is both a maximal projection (merged as complement of V) and a minimal projection (undergoing head movement). Alternatively, it may involve DP movement (phrasal movement), followed by morphological m-merge, as in Matushansky (2006). Since there is no independent evidence of any D or DP movement parameter between Italian and French, the order asymmetries that I have just illustrated do not seem to be predicted. ⁴

Another way of addressing the asymmetry between phrasal and clitic syntax is to call morphology into play. The general idea is that syntactic cliticization may very well produce identical structures in Italian and in French (in fact, in any language), but a morphological component applies its own rules masking this underlying syntactic identity. Here the problem is that a morphological component capable of (re)ordering the clitic string must be quite powerful and also redundant with respect to syntax. Indeed, this question of redundancy, or of Occam’s syntactic razor, in the sense of Arregi & Nevins (2018), is independently discussed in the literature as an undesirable feature of morphological components of the type of Distributed Morphology (Manzini & Savoia 2007; Kayne 2010; Collins & Kayne 2020). ⁵

The alternative to the movement theory is base generation of clitics, specifically as implemented by Sportiche (1996), see (1b). In principle, Sportiche’s approach provides a solution to the order issue, as we fully expect that Voice heads will have their own internal order in the functional sequence and will be able to determine Agree with pro as long as Match and Minimality conditions are met. Ideally, however, one would want to find that clitics project not ad hoc positions, but rather positions that are independently motivated in the functional spine of the verb. This idea is far from new. Within a movement framework, Roberts (2010) has the accusative clitic move to v, on the basis of the fact that the label of the clitic, namely φ, is part of the label of v (as a phase head, i.e. v*). The dative clitic is treated as the head of Appl projections by Cuervo (2003, 2020). In turn, subject clitics have been identified with AgrS (or φP immediately superordinate to IP in the framework of Chomsky 2015) since at least Poletto (2000). What is lacking, however, is an account of the entire clitic string on this basis, within a framework where clitics are first merged as heads.

The central question raised by Sportiche’s (1996) analysis concerns the need for the empty category pro. This empty category was introduced into the theory in order to allow null subject languages to satisfy the EPP. Since then, various scholars have noted the ad hoc character of pro and have proposed that a rich verb inflection of the type that enables null subjects should be allowed to satisfy the EPP by itself (Alexiadou & Anagnostopoulou 1998). Chomsky (2015) provides a more general solution, namely that ‘Italian (and null subject languages generally) … lack the EPP’. If so, pro is required only to allow theta roles to be satisfied by a DP at first merge, independently of whether the overt exponence of the argument is a clitic or even an affix. In other words, pro is a notation for the postulate that theta roles are satisfied uniformly by predicate–argument configurations in phrasal syntax. I consider this Uniformity postulate in the next section.

3. The basic model: accusative clitics and the uniformity principle

Let me begin by considering accusative clitics (Acc). By that label, I mean 3rd person accusative clitics; as I will show later (Section 5.1) there are no accusative 1st and 2nd person (1/2P) clitics in the Romance languages that I considered. ⁶ Acc clitics, as in (12a), consist of a D segment (deictic/definite), namely l, and of a segment carrying gender and number properties, which for present purposes I label φ, as in (12b), cf. Harris (1994).

Structure (12b) is consistent with the thesis in (2b) that clitics are labelled by their intrinsic properties. The next question is whether we can maintain also that they are merged as heads (2a) and that they project their label on the functional spine (2c). Consider example (13a). Following essentially the same steps as in the derivation of a phrasal accusative structure in Chomsky (2015: 14), we obtain the structure in (13b), which fulfils the desiderata. The (verbal) root is merged first; let’s assume that the root is construed as maximal, allowing the asymmetric merger of the clitic head. Merger of v* and inheritance of φ features by R/V result in labelling of the [Cl VP] constituent as φP, namely by the φ features that V and the clitic share.

The crucial aspect of the derivation in (13) that still remains open is whether we can assume that first merge of a clitic as a head yields a well-formed theta configuration, cf. the desideratum in (4). Chomsky (2000: 93) embraces the configurational conception of θ-roles of Hale & Keyser (1993). As a consequence, ‘the θ-Criterion cannot be satisfied … by raising of “θ-features” (the existence of such features aside … feature movement may not be possible)’ (Chomsky 2000: 143). However, there are other alternatives in the literature besides θ configurations and θ features. For instance, Higginbotham (1985) assumes that an argument slot (the so-called R role) is open at N, i.e. the root of an NP, which he assumes to be a predicate. The slot remains open as the NP is variously modified and is ultimately satisfied by another head, namely D.

Therefore, there are at least two paths towards the satisfaction of the Theta Criterion. Argument DPs may be first merged in a theta configuration, i.e. sister configuration with a predicative head, as in Chomsky (1995). Alternatively, argument Ds (articles or clitics) may satisfy the Theta Criterion by closing an open XP (i.e. a predicate) again under sisterhood and at first merge. Note that XP need not be an immediate projection of the predicate head. Consider, for instance, French in (14a). Anticipating the conclusions of the next section, I take the dative clitic to project an Appl category, as in structure (14b). In (14b), Acc is sister to ApplP, therefore neither to the predicate head, nor to its projection (RP/VP). Theta-closure is therefore achieved by predication between D and (still open) ApplP.

Next, if there are two ways of satisfying the internal theta role, by sisterhood with X and by sisterhood with XP, there must also be some principle regulating the traffic. I propose that the relevant principle may simply be Earliness, first suggested by Pesetsky (1989), which I formulate as in (15). Earliness is a form of Economy. Specifically, I suggest that discharging argument slots as early as possible in the derivation contributes to diminish the complexity of the Work Space, in the sense of Chomsky (2020). Therefore, early merge of an argument DP as sister of V is forced for the simple reason that it is possible.

The line of argumentation pursued in (13)–(14) denies the postulate of Uniformity at the syntax/semantics interface, namely that ‘the same meaning always maps onto the same syntactic structure’ (Culicover & Jackendoff 2005: 6, who argue against it), as embodied, for instance, by the Uniformity of Theta Assignment Hypothesis (UTAH) of Baker (1988). Indeed, from a strictly empirical point of view, there are many phenomena where the same semantics has different externalizations cross-linguistically, and also within the same language. To take just one example, in Greek and Albanian, middle-passive voice can be externalized by a periphrasis (be + perfect participle, as in English) or by a specialized inflection or by a pronominal clitic (like Romance se), see Rivero (1990) and Manzini, Roussou & Savoia (2016). Under Uniformity, this kind of evidence is treated by assuming that syntactic structures are underlyingly identical, though they are differently externalized by the lexicon and the morphology of the language. Alternatively, we may assume that middle-passive voice may be realized by different syntactic structures (no Uniformity), provided they all converge on requiring movement of the internal argument to subject position (or whatever the core syntactico-semantic property of middle-passives is taken to be).

I take the view that the issue of Uniformity is ultimately empirical and cannot be decided a priori (see Note 7 for another case in point). Therefore, violation of Uniformity is not in itself a problem for my analysis.

4. The internal order of the clitic string

4.1 Clitic/phrasal asymmetries: Dative

Let us consider the simple clitic cluster involving Acc together with a 3rd person dative clitic (henceforth Dat). Like Acc clitics, Dat clitics have a D lexical base l- followed by gender and number inflections. Thus standard Italian opposes masculine singular gli ‘to him’ to feminine singular le ‘to her’; French has singular lui ‘to him/her’ and plural leur ‘to them’. In addition, Dat clitics also have an oblique case. For instance, French singular lu-i and Italian masculine singular gl-i have a specialized -i oblique inflection, labelled Appl (Applicative) in (16). The dative clitic as a whole is labelled by this Appl (i.e. oblique case) property.

As already mentioned, in French Acc precedes Dat. The order of the two clitics is independent from the interpretation received by Dat, namely, whether it is a selected argument (a goal) as with the verb give or an unselected argument (a benefactive) as with the verb paint, as illustrated in (17). If the Dat clitic is an Appl head, then its position below Acc in French implies that there is an ApplP sandwiched between the Acc clitic (φP) and the root (VP), as in (17b).

Alternatively, in many Romance languages, the order is Dat–Acc, e.g. in Italian as shown in (18). As already discussed in Section 2, the interpretation does not vary with respect to French (17). By comparison with the French structure in (17b), it is natural to propose that Italian has the merger structure in (18b), with ApplP sandwiched between the phase head v* and φP hosting the Acc clitic.

Taken together, (17) and (18) lead us to conclude that in general, Appl is merged freely with respect to φ, though any given language appears to choose one or the other position. I return to this parameter in (41).

The Appl positions now postulated for Romance clitics raise the question as to their relation with the Appl positions introduced by the literature originating with Pylkkänen (2008) for phrasal arguments. According to Pylkkänen, there are two Appl positions, one below V (low Appl) and one between the external argument and the verb (high Appl). She further associates the low Appl with an interpretive relation between the applicative argument and the theme (e.g. goal), and the high Appl with an interpretive relation between the applicative argument and the event (e.g. beneficiary). The difficulties that arise when trying to apply Pylkkänen’s model to Romance languages are widely recognized in the literature (Cuervo 2020 for a review). In Section 2, I follow Folli & Harley (2006) and Boneh & Nash (2012) in assuming that the fundamental distinction in Romance phrasal syntax is between selected and unselected datives. According to those authors, selected datives are hosted in a small clause complement of the root, along the lines of Kayne (1984), Pesetsky (1995), and Harley (2002). Therefore, the structure of Italian (19a) is as in (19b), where it is the P head, a ‘to’, which establishes the applicative relation (roughly possession) between the theme and the goal.

Unselected datives in turn are merged above VP and lower than the external argument. They can be Specs of the Appl positions postulated for Italian and French clitics. Alternatively, however, if the content of the applicative relation is carried by the preposition a, the unselected dative can simply be adjoined to VP, or φP. ⁷ These various options are left open in (20).

In short, in phrasal syntax selected datives (goals) are merged as complements to V in (19) while unselected datives (benefactives) are adjuncts to VP or Specs of ApplP, as in (20). Clitic heads project Appl functional heads above V – where they can satisfy an argument selected by V as well as an unselected argument. Ultimately, then, the order asymmetries between clitic and phrasal arguments follow from the basic difference between the satisfaction of argument slots in head and phrasal syntax. In turn, this asymmetry depends on the abandonment of Uniformity and on the regulatory mechanism of Earliness (Section 3).

Finally, 1/2P datives in Italian have the same position relative to Acc illustrated above for 3rd person Dat. In French, however, Acc, which precedes the 3rd person Dat in (17), follows the 1st/2nd person dative clitic, as in (21). The latter can be accounted for by associating it with the same Appl position as the Italian Dat, as in (22).

The split between (17) and (22) does not depend on the interpretation of the dative in relation to the event, but only on the intrinsic referential properties of the clitic (1/2P vs 3P). Hence the order of clitics is sensitive to the inherent referential content of the clitics, though it is not sensitive to their relation to the event (e.g. goal vs beneficiary). ⁸

4.2 Other obliques

Italian and French have been chosen for exemplification purposes in part because they have a particularly rich clitic set, also including a clitic ne (It.) or en (Fr.) with genitive (Gen) properties, and a clitic ci (It.) or y (Fr.) with instrumental (Ins) properties. Both can be used as locatives (source and location/goal respectively). As the locative meaning is especially salient with ci/y I will normally refer to it as Loc.

Let us then consider the Loc clitic. I assume that it is an Appl head, where the Appl content (Loc/Ins) is projected by the lexical base y/c-. In any given language, the order of Loc with respect to Acc is predictable from the order of Dat and Acc. Thus it is Acc–Loc in French, shown in (23a), and Loc–Acc in Italian, shown in (23b). The two orders follow from the structures in (24).

Suppose then we combine two different applicatives, Dat and Loc. Their relative order is the same in French and in Italian, namely Dat precedes Loc, as in (25). As we fully expect by now, this order is not influenced by the argument roles of the clitics. With a verb like buy the locative is unselected and the dative is a goal; with a verb like put, the locative is the selected goal and the dative is a beneficiary. Yet this is irrelevant for clitic order.

The structures corresponding to (25) are as in (26), where I notate Appl nodes as belonging to different flavours, namely Dat and Loc. Some referentiality hierarchy again governs the relative order of Appls. Thus datives are generally animate while locations are generally inanimate. Animate versus inanimate is a classical split in referentiality hierarchies (or the D hierarchy, in the sense of Kiparsky 2008), where animates are higher than inanimates. Correspondingly, Dat is higher than Loc in the structure.

One last oblique clitic remains to be considered, namely en/ne, the Gen clitic. In French, Gen can combine with Acc, which it follows, as in (27a). Gen also follows other oblique/Appl clitics, as in (27b). I assume that the lexical base en bears the categorial label Appl_Gen, a possible flavour of Appl. ⁹ Since the Gen clitic is lower than Acc, it is merged in the lower Appl positions where other oblique clitics are also merged in French. The structures for (27) are therefore as in (28). If the relative position of Appl clitics is sensitive to the referential (D) hierarchy, we fully expect that Gen is lower than Dat, since Dat is animate. Relative order with respect to Loc may be determined by the fact that Loc is definite. Gen is not necessarily definite either.

In Italian Acc and Gen (ne) are mutually exclusive, as in (29a). ¹⁰ Nevertheless, in Italian as in French, Gen is last in a sequence of oblique clitics, as in (29b), and a potential candidate for the lower Appl position, cf. (29c). ¹¹

At this point, matters are complicated by the fact that en/ne can be construed not only a genitive argument of the verb, as in (27)–(29), but also as a genitive modifier of a DP argument of the verb, for instance in (30a–b). Therefore, the first question that arises under the present approach is how an en/ne clitic that is not an argument of the verb can be merged on the verbal spine. What is more, occurrences of adnominal Gen ne can modify DP objects (30a) or postverbal intransitive subjects (30b); they cannot modify DPs embedded under a PP (30c) or preverbal subject DPs (30d). Belletti & Rizzi (1981) discuss a similar distribution for another type of adnominal ne, namely the so-called partitive, and account for it in terms of government and Subjacency conditions on ne extraction. Though I do not know of any systematic reworking of their proposal in minimalist terms, it is safe to surmise that (30) depicts a distribution accounted for by locality principles. Again, this may be taken to favour a movement derivation for the clitic.

In reality, the case for ne-cliticization in terms of movement is not clear cut, given that, among the various PP complements and modifiers of N, only ne can undergo cliticization. In fact, even the ne clitic in a locative function is excluded, as in (31). In other words, adnominal modifier clitics are restricted to possessors. Given that the generalization opposing possessors to other arguments/modifiers of N is known to hold for movement in general (see especially Cinque 1990 for Italian examples), this may again be taken to support a movement analysis of clitics. The problem, however, is that the generalization itself does not automatically follow from locality conditions such as Minimality or the Phase Impenetrability Condition (PIC).

The special status of possessors points to existence of structural configurations available to possessors, but not to other adnominal complement/modifiers. Szabolcsi (1983) provides classical evidence as to the existence of ‘escape hatches’ dedicated for possessors in Hungarian. Here I will follow another line of explanation, suggested for the partitive ne by Cardinaletti & Giusti (1992) (cf. Cardinaletti & Giusti 2006, 2017) – namely that ne and DP are in a predication relation. It is this relation that is constrained by locality, yielding paradigms of the type in (30) (as assumed by Cardinaletti and Giusti).

Specifically, I assume that the syntax outlined in (28)–(29) for sentential en/ne extends to adnominal en/ne, yielding the structure in (32) for example (30a). Recall next that in many languages possessors display agreement with the N(P) that they modify, essentially like adjectives do. Thus, in Romance, but also in German, possessive pronouns have adjective-like morphology; in Eastern Romance, even phrasal genitives are embedded in so-called linker structures – see Philip (2012) and Franco, Manzini & Savoia (2015) for an agreement treatment of linkers; and Dobrovie-Sorin & Giurgea (2011) and Manzini & Savoia (2018) on Eastern Romance possessives. On this basis, I take the relation between ne and the DP in (32) to be simply Agree.

In turn, the distribution of ne in (30) follows from locality conditions on Agree. Technically, I assume that in addition to its own intrinsic features, ne has features probing for a goal (which must be independently assumed for agreeing pronominal possessors in Romance). An internal argument DP is in the search space of these features, as in (30a–b). However, an external/argument subject is not, nor is a DP embedded under a DP, as in (30c–d). As already mentioned, adnominal ne and its distribution is often considered by the literature in relation to so-called partitive structures. I assume that (32) accounts for Part as well – though I cannot discuss it here for reasons of space. In fact, agreement between ne and DP (or NP) is independently proposed for partitive ne by Belletti & Rizzi (1996).

4.3 Subject clitics

Subject clitics (SCl) and object clitics (OCls) are generally discussed separately in the Romance literature. The reason ultimately goes back to Kayne (1975), who argues for a movement derivation of OCls, but treats French SCls as merely a matter of phonology (or perhaps better in current term, of morphology, i.e. m-merger). As for North Italian SCls, Rizzi (1986) holds that they are inflections and that North Italian varieties are null subject languages (while French is a non-null subject language). Poletto (2000) notes that North Italian SCls cannot be inflections, but nevertheless contrasts OCls derived by movement with SCls treated as functional heads, maintaining the distinction between North Italian and French with respect to the null subject parameter.

In present terms, since I have treated OCls as pure heads, I expect SCls to pattern like them. I also do not expect any difference between French and North Italian, i.e. both are non-null subject languages – see Manzini and Savoia (2005) for similar conclusions. Given the weight of contrary literature, I, however, provide a brief review of the relevant evidence. According to Rizzi (1986), one difference between North Italian and French SCls is that the former but not the latter can be doubled by a lexical DP, as in (33). When it comes to French, however, it is important to distinguish so-called Standard French, i.e. effectively normative French, and Colloquial French. Data in Colloquial French strongly support a parallelism with North Italian varieties. Specifically, Culbertson (2010:118) reports experimental and corpus data converging on the conclusion that ‘Colloquial French shows a high frequency of subject doubling’. Though this is not the place to enter into sociolinguistic distinctions, it is worth pointing out that Colloquial French provides a more congruous comparison with the naturalistic productions obtained by dialectological surveys on North Italian varieties than Standard French.

Another piece of evidence mentioned by Rizzi (1986) is that, in French, the Negation (Neg) clitic follows SCls, while in some North Italian varieties, Neg precedes SCls, as shown for the variety of Chioggia in (34b). However, the Chioggia variety is like French in that SCls can be in complementary distribution with a lexical subject, as in (34a) (see Manzini & Savoia 2005:(I), 47, 131–132, for additional data). Therefore, the expected correlation between parameter values does not hold. Moreover, Culbertson (2010: 96) reports corpus data highlighting interactions between subject clitics and negation also in Colloquial French. In particular, the ne negative clitic is mostly retained when only a DP subject is present, but mostly dropped where a subject clitic is present. ¹²

A third test offered by the literature is coordination. Kayne (1975) observes that in French OCls must be repeated in front of the verb in instances of coordination, as shown in (35a), but this is not the case for SCls, as shown in (35b). In North Italian varieties, SCls are generally repeated, for instance, in (36). Nevertheless, in the Colloquial French corpora of Culbertson (2010: 102), SCls are also mostly repeated in conjunctions.

Overall, I take it that the evidence is not strong enough to stand in the way of treating French like North Italian and SCls like OCls – i.e. in present terms as pure heads.

From the point of view of their internal structure, SCls parallel Acc clitics. They consist of a lexical base which is recognizably the same as that of Acc clitics, namely l- (il- etc.) and of a φ-features inflection, for instance, plural -s in (37a), cf. French ils ‘they’. The 1/2P SCls have essentially the same structure, with Person reference encoded in D (e.g. j in French je ‘I’), as shown in (37b).

Recall further that Acc clitics are merged as functional heads between the V root and the phase head v*, according to the schema in (38).

I propose that there is in fact a single difference between Acc and SCls, namely that SCls are first merged not in the v* phase, but in the C-I phase. Thus, the structure in (39) for SCls parallels the structure provided in (38) for Acc clitics. Following inheritance from C to I, the constituent projected by the SCl head is φP, labelled by the φ features shared by I and the SCl under Agree.

The main difficulty in entertaining a pure head syntax for SCls has also been discussed in relation with Acc clitics, namely the way in which theta roles are satisfied by clitics. In Section 3, I have argued that theta roles can be satisfied by heads, under predication with an open XP. In this respect, I assume that the same again holds for SCls as for Acc clitics – just as long as the embedded V remains in the Work Space (under the PIC). When SCls serve as expletives, they are construed as pure means to satisfy the EPP, with no further interpretive import.

4.4 Intermediate summary: Order

Summarizing so far, clitics are accounted for as pure heads in Romance, and perhaps cross-linguistically. Phrasal (DP) and head (clitic) merger can both satisfy theta relations, but not in a uniform way – as discussed in Section 3 for object clitics and in the previous section for subject clitics. Empirically, as discussed in Section 2, there is considerable evidence that phrasal and clitic arguments have non-uniform syntactic behaviours. The discussion in Section 4 has been devoted to arguing that order asymmetries between phrasal arguments and clitic arguments follow from the fact that the structure of arguments at first merge is strictly determined by their role in the event; thus selected datives/goals are structurally lower than benefactives/unselected datives (Section 4.1). Instead, this event sensitive order is irrelevant for clitic heads. Therefore in French we find both benefactives and locatives (Appls) ordered lower than themes (Acc) – while the reverse holds in Italian.

As for what actually does order clitics, we have seen that there is no constraint as to how the accusative clitic is ordered with respect to applicatives (see the French/Italian parameter), but on the contrary the internal order of Appls tends to be fixed. Both facts are surprising when seen in the light of phrasal order, since the order of phrasal objects tends to be fixed, while the relative order of phrasal adjuncts may be quite free. Let me consider the relative order of Appl clitics first. In the course of the discussion, I have suggested that Appl clitic order seems to respond to Person, animacy, and definiteness hierarchies. I follow Kiparsky (2008) in assuming that there is in fact a single referentiality or D-hierarchy, for which I adopt the formulation in (40).

To the extent that an order of merger of Appl clitics can be discerned, it is from lower to higher in the hierarchy. Thus Gen, which is the only indefinite clitic in the system, capable of referring in particular to mass terms in partitive uses, is the lowest Appl in both Italian and French. In turn, Loc (inanimate) is higher than Gen, but lower than Dat (animate). In French, there is a single Appl higher than Acc, and it is reserved for 1/2P. The plausibility of the proposed generalization is strengthened by the observation that clitics are independently known to be sensitive to referential prominence constraints such as the Person Case Constraint (PCC, Bonet 1991), whereas phrasal syntax is generally immune from them. I suggest that these two sets of data may be related. ¹³

The last point to be discussed concerns the relative order of Acc and Dat or, more generally, of Acc and Appls. While the relative order of Appls is not different in the two languages, French (or Corsican, see Note 8) attaches Appls lower than Acc and Italian higher. This parametric choice is schematized in (41).

Finer parameters may also be instantiated. Under the French person split, the Acc clitic precedes all clitics with which it combines, except 1/2P clitics. ¹⁴ The parameter in (41) can take this into account with a minimal modification, as in (41′).

The real issue is why there would be a parameter like (41). I believe the answer is essentially the one expected under current minimalist theories (Chomsky 2013; Chomsky, Gallego & Ott 2019), namely that (41) does not represent a hard-wired switch of some sort. Rather the choice arises simply because the grammar fails to impose any order constraint on the syntactic objects involved. In phrasal syntax, merger of themes before applicatives is forced by the configurational definition of theta roles. But for reasons that are independent of the parameter in (41), I have concluded in favour of a predicational satisfaction of theta roles by clitics. At this point, then, the grammar does not decide the relative order of Appls and φ, and either order can be found.

5. Phrasal/clitic asymmetries: case and agreement

So far, I have focussed mainly on the relative order of clitics and on the fact that it differs from the order of their supposed phrasal counterparts, arguing for an account of clitic order, based on their pure head status. If this approach is correct, we expect other syntactic asymmetries to be found between clitics and phrasal arguments. Such asymmetries are indeed found in the domain of case and agreement. In Sections 5.1–Section 5.2, I address case asymmetries in the form of Differential Object Marking (DOM). In so doing, I also consider clitic doubling, completing a piece of the clitic puzzle missing in the discussion so far. In Section 5.3, I review classical evidence that clitics but not phrasal arguments agree with perfect participles in Italian and in French (Kayne 1989).

5.1 Differential object marking of clitics

So far, I have not considered accusative 1/2P clitics. These are morphologically syncretic with 1/2P dative clitics. More importantly, the position of 1/2P clitics does not vary in relation to their accusative or dative status, as can be seen in Italian. The 1/2P clitic precedes Loc in a dative function, as in (42a), and in an accusative function, as in (42b), contrasting with 3Acc, which follows Loc, cf. (23b).

There is another phenomenon in Romance, where referentially high-ranked direct objects show up with dative morphology, namely Differential Object Marking (DOM). For instance, in Spanish in (43a), the high-ranked direct object surfaces not as a bare DP accusative, but as a PP headed by the dative preposition a. Both the descriptive and the theoretical literature (Bárány 2018 and references quoted there) generally treat the preposition as a differential mark of accusative. However, Torrego (2010) and Manzini & Franco (2016) argue that under DOM (in Romance) a referentially high-ranked internal argument must in fact be merged as an Appl, and cannot be merged as an accusative theme. Intuitively, the idea is that the argument structure of call someone in DOM languages is obligatorily rendered as make a call to someone. In (43b), I reproduce the structural implementation of this idea by Torrego (2010) for Spanish (slightly modified), where contratar ‘hire’ is essentially ‘do a contract (to)’. ¹⁵

Against this background, I adopt the proposal by Manzini & Savoia (2018) that 1/2P clitics in Romance undergo DOM. It follows that they are merged as Appl heads not only when they are goals/beneficiaries, but also when they are first internal arguments. Both sentences in (42) are then associated with the (partial) structure in (44a). As for the morphological structure of 1/2P clitics, I assume they consist of a lexical base m (t, etc.) denoting Speaker (Hearer, etc.) followed by an Appl inflection, as in (44b). Hence, as I have maintained throughout, the phrasal label (ApplP) is projected from the label of the clitic.

Turning to French, the fact that 1/2P clitics always precede Loc, as in (45a), does not tell us much, since 3Acc clitics also precede Loc. In the absence of positional evidence, we must rely on the fact that 1/2P clitics are syncretic between accusative and dative occurrences as evidence that DOM does apply, as in structure (45b). ¹⁶

Let us then go back to the main theme of the present discussion, namely asymmetries between clitic and phrasal syntax. Unlike Spanish in (43), ¹⁷ French and Italian lack DOM in phrasal syntax. Therefore, 1/2P arguments are treated as ordinary accusatives in phrasal syntax, but as DOM objects in clitic syntax. Generalizing, clitics and phrasal arguments are not only differently ordered by the syntax, but they also may be associated with different case patterns.

In Spanish, where DOM affects phrasal arguments, as in (43), DOM lexical objects are pronominalized/doubled by Acc clitics, as in (46a). This is unexpected under the construal of DOM as an Appl/oblique element adopted here. However, there are also varieties of Spanish that pronominalize or double the DOM object by the Dat clitic le/les ‘to him/her, to them’ (whence the traditional label of leísmo for this pattern). An example from a Basque leísta variety is given in (46b) (Ormazabal & Romero 2013: 316).

In present terms, (46b) is the unremarkable case in which DOM is found in both phrasal and clitic syntax. I propose then to account for (46a) by assuming that clitics and phrasal arguments can also have independent case patterns, as concluded above for 1/2P clitics. In the Spanish examples, however, the clitic may not only pronominalize a direct object, but also double it. Prima facie, then, my proposal seems to allow case disagreement between a clitic and its phrasal double, namely in examples like (46a). In order to consider this issue, I need first to be explicit about my analysis of clitic doubling. I turn to it in the next section, where I will also take up the examples in (46) in more detail.

5.2 Clitic doubling

Consider Spanish in (47a), where the Appl_Dat clitic le doubles a thematic dative a PP. I assume that the PP satisfies both the goal thematic configuration and the dative inherent case, in the sense of Chomsky (1986), i.e. the dative case associated with the goal thematic configuration. As a consequence, the Dat clitic le does not have any theta slot to saturate. I suggest that this is possible insofar as the φ features of the clitic are uninterpretable. But precisely because they are uninterpretable, they act as a probe, triggering Agree and taking the a PP as their goal, along the lines of (47b). ¹⁸

The analysis proposed for (47) can be generalized to all instances of clitic doubling. Clitics have two possibilities when they merge. One possibility is the one that I have investigated before in this article, namely that they saturate an open slot in the predicate and receive an interpretation as arguments of that predicate; this requires them to be referential, hence to have interpretable φ-features. The other possibility is that upon merging they do not saturate a theta role. This means that their φ-features are uninterpretable and probe for a goal, represented by some argument DP/PP. ¹⁹ An analysis of this type in (47) can also be applied to subject clitic doubling. Thus, a clitic in complementary distribution with a lexical subject is an argument with interpretable φ-features saturating a thematic role. A clitic co-occurring with a lexical subject is a probe (uninterpretable φ-features) whose goal is the lexical subject. ²⁰

Assuming that the possibility of clitic doubling is analysed as just proposed, we still need to explain why clitic doubling is necessary in some cases. The key observation in the literature on object clitic doubling is that it depends on certain appropriate semantico-pragmatic conditions; for the considerable range of proposals found in the literature, I refer the reader to Anagnostopoulou (2017). Concretely, in Ibero-Romance varieties, arguments that undergo clitic doubling vary along the lines of (48) (Assmann 2017) (DO abbreviates direct objects, IO indirect objects).

It is evident that the notions of deixis, animacy, and specificity in (48) refer back to the Referentiality hierarchy in (40), which also governs DOM. Because of this, Kallulli (2016) and Kiss (2017) explicitly suggest that clitic doubling marks the same kind of referential prominence that is also marked by differential case. I tentatively assume that an explanation along these lines is correct (though see Manzini & Savoia 2018 for specific criticisms of Kallulli). In other words, differential marking of the object (DOM) can be achieved by special agreement morphology (the clitic) on the V spine, i.e. by head marking in the sense of Nichols (1992) as well as by dependent marking, i.e. special case morphology on the object. The prevalence of IO doubling over DO doubling can perhaps be given a unified explanation with the preference for referentially high ranking DO referents, keeping in mind the properties of datives as logophoric centres (Pancheva & Zubizarreta 2018 and references quoted therein). ²¹

With this background on clitic doubling, we may then go back to the example set in (46). Standard Spanish (46a) is often taken as an argument that the a phrase is a ‘prepositional accusative’ and cannot be a dative, since it is doubled by an Acc clitic. By the same token, however, the leísta dialects like (46b) are problematic. I propose to account for both examples in (46) on the basis of a classical idea in the generative treatment of oblique arguments, namely that they are in fact ambiguous between the PP and the DP status (Selkirk 1977; Pesetsky 1982 on pseudopartitives; Manzini & Franco 2019 for a recent survey). The natural construal of this idea is in terms of labelling. In other words, in a [[P] [DP]] structure, P does not necessarily label the resulting constituent as PP; rather the whole constituent may be labelled as DP by the D head of DP.

Consider then (46a), repeated as (49a). I assume that the referentially high ranked argument el is merged with the Appl/P head a. Following the literature quoted, two different labelling possibilities are open at this point. One is that the Appl/P head is adjoined to DP as a case modifier, so that the whole a el constituent is labelled as DP, as in (49b). This DP is then probed for by an Acc clitic, with which it agrees in φ features. The resulting pattern is that attested in standard Spanish.

Consider next the leísta example in (46b), repeated as (50a). In leísta dialects, I assume that the a preposition embeds DP (here ‘the boy’) as its complement, so that the whole constituent al niño is labelled as PP by the P head, as in (50). As a result, it is matched by an Appl_Dat clitic, namely le.

In short, DOM arguments can project either as PP or as DP, and will be doubled by Acc or Dat clitics accordingly, as in (49) and (50) respectively. The choice between DP and PP represents the parameter between standard and leísta dialects respectively. One may therefore wonder why the structure of the goal dative in (47a) is always (47b), with the goal dative projecting PP. Following Manzini & Franco (2016), I take it that this and other differences between DOM and goal datives follow from the fact that DOM datives are structural obliques, determined by a configuration involving a high ranked internal argument. On the contrary, goal datives are inherent obliques, selected by the verb. Their selected status imposes projecting the PP label, selected by the verb.

In fact, the patterns considered so far do not exhaust known variation. In South Italian varieties, Acc clitics pronominalize both theme and goal arguments, as in (51) (Manzini & Savoia 2005). This pattern is also known in Spanish varieties, especially of Latin America (Zdrojewski & Sánchez 2014) and is referred to as loísmo/laísmo in the Spanish descriptive tradition.

The labelling mechanisms proposed in the text suggests that in South Italian varieties a phrases project as DPs independently of whether they are structural or inherent datives. In other words, the contrasting treatment of structural and inherent obliques is itself open to variation.

5.3 Perfect participle agreement

The last asymmetry between phrases and clitics to be considered here concerns perfect participles. In Italian and French, Acc clitics agree with perfect participles as in (52a), but lexical object DPs do not, as in (52b).

Under Kayne’s (1989) movement analysis, clitics agree with the participle in that their movement path (from argument position to IP-adjoined position) passes through the Spec of an agreement head, namely AgrO. The Spec-head configuration triggers agreement with the AgrO head, hence with the perfect participle. By contrast, a lexical object, remaining in situ, does not enter into the crucial Spec-head configuration with AgrO and does not trigger agreement. Under minimalist assumptions, however, the asymmetry between clitics and lexical DPs is lost in crucial respects. In standard minimalism, lexical DPs need to enter Agree with the v* head for case purposes. Therefore, the contrast between agreeing clitics and non-agreeing DPs can no longer be imputed to the fact that the former move to AgrO/v* entering agreement – and the latter don’t.

D’Alessandro & Roberts (2008) propose an alternative analysis of the pattern in (52) in terms of externalization. For them, the externalization of Agree relations takes place only if probe and goal are within the same spell-out domain, as defined by Chomsky’s (2001) Phase Impenetrability Condition (PIC). Specifically, when the phase head C-I is introduced, the VP complement of v is spelled out. At that point the participle is no longer in V (because they assume it has been moved to v), and no agreement with a lexical DP in situ (i.e. within VP) can be externalized. On the contrary, the clitic head is adjoined to the v head (also hosting the participle) – so that the clitic and the participle are in the same phasal spell-out domain, and display agreement.

The asymmetry introduced here between phrasal arguments and clitics as pure heads suggests a different treatment of (52). Under the assumptions first introduced in Section 3 (based on Chomsky 2015), the lexical verb is merged in the root position, followed by merger of the clitic, merger of the v* phase head, inheritance from v* to the root/V, and labelling of φP. Note, however, that the clitic intervenes on the (v, V) inheritance path. I propose that the result may be a complex inheritance sequence from v* to the clitic and to V, as schematized in (53).

Consider next the lexical object in (52b). Italian has no EPP of the v* phase and therefore no obligatory raising of DP to φP, in the sense of Chomsky (2015). Thus in (54) merger of the participle with the internal argument is followed by inheritance from v* to V and labelling of VP. V enters Agree with DP, not as part of the head-to-head inheritance sequence, but as an independent feature checking relation. I therefore suggest that while head-to-head inheritance in (53) is externalized as overt agreement of the clitic and the V head, feature checking in (54) between the V head and the phrasal object has no overt reflex. ²²

Interestingly, many Romance languages present no asymmetry between clitic and phrasal objects. In some Central and South Italian varieties the participle agrees with both phrasal and clitic direct objects as in (55a) (Manzini & Savoia 2005: §5.1.2, §5.6.1; D’Alessandro & Roberts 2008, 2010 for additional evidence). Vice versa, in languages like Spanish, neither the clitic nor a phrasal object triggers any agreement with the perfect participle, as in (55b).

In present terms, Spanish and Salentine have the same underlying grammar as Italian and French. They differ from those languages in that, at externalization, no agreement relation with v* has an overt reflex (Spanish), or all agreement relations with v* are overtly realized (Salentine). ²³

6. Conclusions and further problems

Asymmetries between clitics and phrasal arguments have long been known, concerning, among others, their order (with respect to one another), their case (e.g. DOM), and their agreement (e.g. perfect participle agreement). I have argued that this asymmetry is syntactic, involving the EXT procedure, but not any post-syntactic morphological component. Specifically, I have argued that clitics have a pure head syntax. This means that while clitic and full pronouns converge at the SEM(antic) interface, they are associated with different thematic structures (Section 3). Phrasal arguments are ordered by their eventive role; clitics are ordered by their intrinsic referential features (Section 4). Clitics also admit of a non-argumental construal (in some languages), seen in clitic doubling (Section 5).

Several issues open as a consequence of the proposals put forth here. To begin with, some form of incorporation of clitics with the verbal head is generally deemed necessary to yield the typical behaviour of clitics as forming a single unit with the verb. However, incorporation is undesirable in present terms, since it reintroduces operations of head movement that are otherwise avoided in the generation of the clitic string. A promising alternative approached is proposed by Manzini & Pescarini (2021), whose key assumption is that clitic heads are not introduced into the derivation by Set Merge, but rather by Pair Merge. Specifically <Cl, H> merge pairs, where H is a verbal head (say v) are built into Pair Merge sequences, in the sense of Chomsky (2020). By construction, no member of the sequence can be singled out by grammatical operations – which is the essence of incorporation. Crucially, Manzini and Pescarini’s (2021) model presupposes that clitics are pure heads, as argued here.

Furthermore, only in a few Romance languages can the clitic string be seen in the v phase where it is first merged according to the model developed here, notably varieties of north-east Piedmont like Borgomanero (Tortora 2014; Manzini & Savoia 2005). In most modern Romance languages, the clitic string is associated with the I area of the sentence (Kayne 1991, Pescarini 2021 for an exhaustive picture of attested variation). The literature generally accounts for these facts in terms of head movement (Kayne 1991; Roberts 2010) – which is undesirable here for the same reasons as incorporation above. Manzini & Pescarini (2021) again offer an alternative based on the idea that clitic strings correspond to Pair Merge sequences. Specifically, they propose that the Pair Merge sequence can be built with any phase head as a Link (in the sense of Chomsky 2020) – hence with v or with (C-)I. In short, it may be possible to avoid recourse to head movement altogether in the derivation of clitic structures.