A syntactic account of auxiliary selection in French

2.1 HAVE-selecting verbs

As noted in the introduction already, non-reflexive verbs that select an EA form compound tenses with have in French. This category includes all unergatives (2), transitives (3), and ditransitives (4).

Past participle agreement morphology is found under a specific condition, namely when the IA linearly precedes the verb, whether it is a clitic (5) or a pre-posed DP (6).

As is well known, past participle agreement morphology tends to be optionally realised in colloquial French (Audibert-Gibier 1992; Blanche-Benveniste 2006; Georgi and Stark 2020; MacKenzie 2013; Stark 2015). It is silent on most verbs, where it is only present in orthography, and only gender agreement can have a morphophonological realisation. There is, however, a phonological condition: as illustrated in examples (5) and (6), the participle must end in an underlying obstruent for agreement to be overt. An anonymous reviewer points out, however, that in Swiss French, agreement is realised as an offglide [j] after a vowel, making it much more audibly frequent.^[3] They give the following examples and transcriptions for aimée (7a) and vue (7b).

This brief overview concludes the distribution of have and its link to past participle agreement morphology, for which I review formal accounts in Section 3.2. Let us now turn to the distribution of be.

2.2 BE-selecting verbs

Unaccusatives (8) select be, and past participle agreement morphology systematically spells out, independently of the linear order of constituents, as shown with (9).^[4]

Reflexive verbs also pattern with be, whether they select one (10) or two (11) arguments.

Unlike unaccusatives, agreement is only ever found if the IA precedes the past participle (this is similar to the pattern identified with verbs that select have), contrast (10) with (12), and (11) with (13).

The remainder of this paper will refer to the constructions exemplified here, for which it will offer a formal account.

3.1 Previous approaches to auxiliary selection

Auxiliary selection shows significant crosslinguistic variation in Romance (for instance, Schifano, Silvestri & Squillaci 2016 report 8 different patterns, see also Manzini and Savoia 2005 for a thorough documentation across different Romance varieties). The discussion that ensues aims at setting the scene for a formal analysis of auxiliary selection in French.

Bentley and Eythórsson (2004) build on Sorace (2000) in arguing for two semantic classes of unaccusatives: a core class (be is systematically found) and a peripheral class (auxiliary selection is subject to crosslinguistic or crossdialectal variation). By analysing crosslinguistic data, these authors claim that the distribution of have and be does not systematically connect to the presence of absence of an EA, and therefore cannot be determined by syntax. They instead develop an account whereby auxiliary selection is a morphological rule sensitive to the semantics of predicates. Admittedly, Bentley and Eythórsson (2004) face the issue of exceptional cases in Standard French (for instance changer ‘change’ takes have whereas rester ‘stay’ takes be), and they postulate that be in Canadian French is a historical residue subject to frequency. Whilst their proposal makes interesting predictions, the tests they use to assess the semantic connection between a given predicate and its selected auxiliary are all based on Italian examples and do not apply to French. Furthermore, Bentley & Eythórsson’s (2004) semantic analysis does not readily explain auxiliary selection with reflexive verbs and requires the assumption of a morphological [+pronominal] feature that overrides semantic criteria. The motivation for the connection between [+pronominal] and be is not clear and is seemingly ad hoc.

Within a syntactic analysis, Bjorkman (2011) assumes that the auxiliary is the Spell Out of a Perf(ective)-head sandwiched between TP and VP and argues that auxiliary selection results from agreement or lack thereof between Perf and V: be surfaces when Agree holds, whereas the presence of an intervening EA between the two heads blocks Agree and yields have. As a result, auxiliary selection directly connects to the presence or absence of an EA (contra Bentley and Eythórsson 2004). Her proposal includes the view that semantics is involved in the case of peripheral unaccusatives, and she remains agnostic with regards to auxiliary selection with reflexive verbs. As already discussed in Section 2, French showcases predicates that select an EA and form compound tenses with be, therefore this proposal can only account for a subset of verbs.

Building on D’Alessandro and Roberts (2008, 2010) propose an iterated vP (14), where v_Aux spells out the auxiliary and v_Prt the past participle. The V-head incorporates with v_Prt to form the past participle.

(14)

They argue that v_Aux is realised as have or be depending on its set of valued features at Spell Out and intend a generalisation whereby v_Aux surfaces as have if its v_Prt-complement is non-defective (i.e., a v-head is non-defective when it assigns an external θ-role, and probes the direct object’s ϕ–features and licences its Case feature, as proposed by Chomsky 2001: p. 43). As these authors point out, this generalisation does not account for the presence of be with reflexive verbs that are non-defective (i.e., those that select both an EA and an IA).

In a recent series of works on Italo-Romance, Amato (2021, 2022, 2023, 2024) takes the auxiliary to be the Spell Out of Perf, a functional head merged between TP and vP (this structure is not different from the ones proposed by D’Alessandro and Roberts 2008, 2010; Bjorkman 2011). Her proposal relies on the assumption that the allomorph is conditioned by the formal features valued on Perf: the IA’s ϕ-features are probed by v first, enabling Perf to copy the IA’s person values as they appear on v. Given the undesirable effect that the EA acts as an intermediate potential goal, as shown in (15), Amato (2021, 2022) puts forth the mechanism of Nested Agree, which formalises the order in which instances of Agree take place when a given head has more than one feature.^[5]

(15)

Accordingly, as illustrated in (16), Perf is merged with a valued probe Infl that establishes an Agree relation with v (16a).^[6] This operation, she argues, allows Perf to further exploit this channel with v, allowing for [uPers] to probe its value as in (16b). This proposal elegantly captures how the IA’s person features are copied on Perf despite the EA being the closest potential goal.

(16)

a.

b.

Based on this derivation, and in the context of Standard Italian (which shows a pattern comparable to that of Standard French), Amato (2022) argues that have yields where Perf copies v’s inherited person feature (i.e., with transitives and unergatives),^[7] whereas be obtains where Agree fails, assuming that unaccusative v-heads are not ϕ-probes. Nonetheless, this proposal does not capture why we find be with reflexive transitive verbs that select a post-verbal direct object: in this context, v and Perf should Agree, which incorrectly predicts have. In Amato (2023: pp. 65–69), she addresses this issue in postulating that the reflexive clitic is merged in the specifier of an applicative head between v and V (17). In this configuration, the IA is out of reach from v, which probes the reflexive clitic instead. The reflexive, she argues, bears an unvalued person feature when it enters the derivation. Agree stops upon finding the matching unvalued feature, leading to v not obtaining a person value for Perf to probe, and spelling out be as described above.

(17)

Whilst French auxiliary selection is ‘argument-structure-driven’, let us briefly consider ‘person-driven’ systems where the choice of the auxiliary depends on the person features of the subject.^[8] D’Alessandro and Ledgeway (2010) and D’Alessandro and Roberts (2010) analyse this issue in Abruzzese, and they reach the conclusion that be spells out when a higher v-head agrees with the subject, which convincingly suggests that the functional head spelling out the auxiliary is associated with unvalued person features, at least in some varieties. This hypothesis is further developed by D’Alessandro (2017), who introduces structure (18) where the π-head is a ϕ-bundle (the content of which is language-specific) spelling out the auxiliary. Further, this author argues that the (content of the) v-head is scattered between v itself and π, thus forming a complex probe (I return to this notion in Section 4).

Although not articulated in the context of French, this proposal entertains the notion that auxiliary selection interacts with (a subset of) ϕ-features. D’Alessandro (2017) convincingly argues that the Spell Out of the auxiliary-head is dependent on the goal(s) probed by T and/or v, as previously formulated by D’Alessandro and Roberts (2010). As further demonstrated by Amato (2022, 2023), there may be bridges between ‘argument-structure-driven’ and ‘person-driven’ systems.

The common theme identified here centres on the ϕ-features as they appear on the auxiliary, in particular for person. I will show that the same assumption can account for auxiliary selection in French, yet before doing so, let us consider how to account for past participle agreement morphology or lack thereof.

3.2 Syntactic approaches to past participle agreement

The objective of this paper is to offer a novel analysis of auxiliary selection. As exposed in the empirical discussion of Section 2, past participle agreement morphology represents the opposite side of the coin and cannot be ignored. It is treated as a peripheral issue throughout the paper, yet the proposal ensures its accurate realisation. A comprehensive review of the literature is therefore essential, as it contributes to a more nuanced understanding of the syntax of compound tenses and of the featural makeup of the relevant functional heads.

Within an earlier generative framework, Kayne (1989) puts forward a unified theory proposal to account for both finite verb agreement with the subject and past participle agreement with the IA in French and Italian. Given that past participle agreement surfaces when the IA (clitic or full DP) linearly precedes have, Kayne (1989) argues that the functional head Agr taking VP in its complement spells out agreement when the IA transits through its specifier (19b-19c), a proposal that Belletti (2006) adopts and builds on. The specifier of VP remains empty with a post-verbal IA (19a), therefore the past participle does not morphophonologically agree with it. This proposal, however, does not straightforwardly capture why agreement is found on the past participle when the auxiliary is be and the IA post-verbal.

Within a Minimalist framework (and replacing Agr with v), D’Alessandro and Roberts (2008) note that v ϕ-probes the IA in (19a), which incorrectly predicts the Spell Out of agreement morphology. To address this issue, these authors argue that the agreeing elements (i.e., the IA and the past participle) must belong to the same phase at Spell Out for past participle agreement morphology to obtain (prosodic phrases are assumed to derive from Transfer domains, cf. Chomsky 2001). When the predicate is transitive, v_Prt assigns an external θ-role and functions as a phase boundary, therefore past participle agreement is set to default if the IA remains in situ (20a), as the two agreeing elements are in different Transfer domains. When the IA is a clitic (20b), however, agreement is triggered since it lands within the same domain as the past participle (v_Aux is not assumed to ever be phasal).

(20)

a.

b.

Turning to unaccusatives (21), v_Prt does not select an EA. Because its phasehood is deactivated, the IA is spelled out in the same domain as the past participle, its probe. As a result, D’Alessandro & Roberts’ (2008) analysis correctly predicts that past participle agreement morphology occurs in both (20b) and (21).

(21)

This proposal elegantly captures the distribution of past participle agreement morphology with a straightforward explanation. Nevertheless, it does not readily account for the Spell Out of agreement with A′-movement, unless one posits that intermediate traces are taken in consideration (which should not pose a problem, as an anonymous reviewer points out).

Building on D’Alessandro & Roberts’ (2008) hypothesis, Kobayashi (2022) proposes that v (which spells out the participle) bears unvalued features for gender and number only, and that agreement morphology depends on the type of Agree operation in place (i.e., either Minimal Agree or Full Agree). In Minimal Agree, due to minimal computation, the probe targets only the highest head of the phrasal complement, minimally valuing its ϕ-features (e.g., number only). Full Agree, on the other hand, occurs when the probe finds a match for all of its ϕ-features.^[9] Specifically, with a post-verbal IA, the ϕ-probe v searches for matching values within its complement and halts upon encountering the first match: this is Minimal Agree (22).^[10] According to Kobayashi (2022), past participle agreement morphology is realised at Spell Out only when the complex head Prt+v finds matching values for all of its features (i.e., both gender and number). In other words, a minimally valued participle spells out in its default agreement form.

(22)

With a non-phasal v however (i.e., when v_Aux spells out as be), the IA and Prt+v are contained within the same Transfer domain (23). Upon landing of the IA into the specifier of TP, the Spec-Head relation values T’s ϕ-features, and the search continues within the phase, leading to full valuation of Prt+v’s ϕ-features and spelling out past participle agreement morphology in gender and number.^[11]

(23)

Based on the empirical observation that past participle agreement morphology tends to be optionally realised in colloquial French with object clitics, Kobayashi (2022) further argues that Prt+v can agree with either the lower copy of the clitic (i.e., this is the mechanism exposed in (22)), yielding default agreement, or with the moved clitic (similar to how agreement obtains with a moved IA (23)). The structure of past participle agreement morphology or lack thereof with object clitics is given in (24): option 1 yields agreement, whereas option 2 does not.

(24)

In sum, Kobayashi (2022) offers a powerful analysis that correctly predicts the distribution of past participle agreement morphology as found in French.

The asymmetry of past participle agreement morphology between pre-verbal clitics and post-verbal lexical objects is also tackled by Manzini (2023), who departs from the studies mentioned until now in assuming that object clitics are generated higher than full object DPs, namely between vP and VP (25b).

This alternative approach takes clitics to be merged as pure heads: the lexical root V is merged first as a maximal projection, which enables “the asymmetric merger of the clitic head” (Manzini 2023: p. 95). Subsequent merger of the v* phase head triggers inheritance of ϕ-features to the clitic, allowing the [Cl VP] constituent to project a ϕP. With regards to past participle agreement, Manzini (2023) proposes that clitics, as functional heads, enter into an Agree relation with the participle. The latter inherits the clitic’s ϕ-features (following the framework of Chomsky 2015), triggering morphological agreement.^[12] Lexical objects, on the other hand, do not trigger agreement because they remain in situ within the VP; as a result, the participle does not inherit features from the DP, and no agreement is triggered. This hypothesis mainly centres on the specific issue of past participle agreement morphology with transitive verbs.

So far, we have observed a general tendency for authors to rely on Agree in their proposal. An opposing view is developed by Georgi and Stark (2020), who argue against a unified view of past participle agreement. For them, there exist two mechanisms. Firstly, with be, the past participle v is a ϕ-probe that copies the IA’s ϕ-features in situ, yielding obligatorily overt morphology (whether or not the IA satisfies an EPP in the specifier of TP). The second mechanism applies to contexts with have, for which they claim that v is not a ϕ-probe. They argue that agreement morphology is triggered only if the IA is a resumptive pronoun (i.e., a clitic) that incorporates with v.^[13] In other terms, they argue that past participle agreement with be is a product of Agree, while with have it is a result of incorporation. This mechanism enables them to account for the mismatch between obligatory agreement in (26a) and its absence in (26b), which both feature a post-verbal IA: in (26b), the past participle does not have ϕ-features.

This proposal is problematic as it forces the assumption that v is somewhat defective with have, which goes against what most authors believe.^[14] Furthermore, it heavily relies on the assumption that past participle agreement morphology is obligatory with be whether or not the IA is post-verbal, which leaves examples like (27a), where a reflexive clitic is present, unaccounted for (see Section 2 for complementary examples).

While most authors link past participle agreement morphology to auxiliary selection, Bentley and Eythórsson (2004) argue that the two should be treated independently. Their reasoning is based on the observation that in some Romance varieties, the past participle selected by have exhibits morphological agreement with the EA, unlike in Standard French and Standard Italian. However, this claim is primarily based on Italian dialects, where auxiliary selection follows a pattern distinct from that of French (i.e., the so-called ‘person-driven’ systems mentioned in Section 3.1).

Given that agreement morphology inflects for gender and number only, I adopt the assumption that the past participle acts as a probe for these two features exclusively. In so doing, I follow the approaches of D’Alessandro and Roberts (2010) and Kobayashi (2022), which align with Kayne’s (1989) earlier proposal.

6.1 When the internal argument is a post-verbal DP

With a post-verbal lexical IA, we find have and the past participle lacks agreement morphology (38a). The EA is merged in the specifier of v_Prt, which, as we saw earlier, is a head that probes for a V-feature and ϕ-features. The derivation of (38a) is given in (38b).

(38)

a.

Camille	a	pris	la	fleur	(French)
C.	have-prs	take-ptcp	the	flower
‘Camille took the flower’

b.

As usual, v_Prt probes the verb, an operation which triggers V-incorporation, and it also probes the ϕ-features of the IA. As mentioned above in the context of unaccusatives, the person feature of the IA is shared upwards from v_Prt to v_Aux once the complex head starts probing. Note that share offers a simple explanation as to why v_Aux does not consider the EA a potential probe: its person feature is already valued by the IA.^[19]

As opposed to the structure for unaccusative verbs given in (37), we do not obtain a co-indexed chain of identical person features between T and v_Aux here. Instead, the two heads probe different goals: T probes the EA’s ϕ-features whereas v_Aux probes the IA’s person feature. Even if the two heads have a [3-person] value, they do not share the same identity (cf. example (31c) and relevant discussion in Section 4). Following the vocabulary insertion rule given in (32b), the lack of co-indexation between the two forces have-selection.

This analysis also extends to unergatives, where Agree fails on both v_Aux/Prt due to the absence of a goal (there is no IA in this context). As a result, past participle agreement morphology is absent, and have is selected, since v_Aux lacks a person value.^[20]

6.2 When the internal argument is a pre-verbal clitic

As illustrated in Section 2.1, the past participle does not inflect in French when the IA is post-verbal (38a), but it does if the IA ends up in a pre-verbal position. Consider (39), where the IA is a clitic that attaches to the auxiliary and feminine agreement is pronounced.

If my analysis of auxiliary selection is on the right track, be is the Spell Out of a co-indexed chain of person features between v_Aux and T, whereas the auxiliary have is the elsewhere allomorph and is not conditioned by the valuation and co-indexation of ϕ-features (cf. insertion rules given in (32)). Let us consider auxiliary selection with pre-verbal clitics; the structure of (39) is given in (40). The IA is generated as a clitic in the complement position of V (Gallego 2016; Kayne 1975; Olivier 2022; Roberts 2010), and it is subsequently probed by the non-defective v_Prt along the mechanism adopted until now. Being a clitic, the IA moves to the left-edge of v_Prt (cf. Roberts’ 2010 analysis), before being transferred up to v_Aux to which it incorporates. I follow D’Alessandro and Roberts (2008) and Kobayashi (2022) on the claim that past participle agreement morphology spells out when the IA and the past participle are in the same Transfer domain.

(40)

The derivation of structure (40) goes as follows: v_Prt probes V to value its V-feature, before probing the clitic and subsequently transferring it to v_Aux – this operation is essentially a form of clitic climbing, which is itself a byproduct of share.^[21] This analysis therefore unifies climbing of the IA to v_Aux and past participle agreement morphology on v_Aux: they are both triggered by the same operation (i.e., ϕ-probing of the IA by a complex head), and they are sent to Spell Out together. Finally, the complex [IA_clitic+v_Aux] incorporates with T, and the EA moves to the specifier of TP to satisfy the EPP and value its ϕ-features. Similar to when the IA remains in situ (38b), v_Aux and T are valued by different goals, therefore they do not form a co-indexed chain of identical person features – in contrast to the earlier structure for unaccusatives (37). Following the proposal developed here, have is correctly predicted.

A key consequence of this proposal is that the valuation of v_Aux’s person feature results in the Spell Out of be only if this very feature is also probed by T and encodes the same identity (cf. discussion with example (31) and insertion rules in (32)). Thus, I argue against the traditional hypothesis that auxiliary selection is sensitive to argument structure. Building on this, I address auxiliary selection with reflexive verbs in the following section.

7.1 The issue

Firstly, the analysis proposed for unaccusative verbs in Section 5 can be extended to reciprocals (41), inchoatives (42), and middle passives (43). Prima facie, these predicates appear to be transitive (i.e., the reflexive clitic looks like an argument). However, I adopt the assumption that they trigger movement of the IA to the subject position, whereas the reflexive clitic absorbs both the external θ-role and the ability to assign accusative case (Belletti 1982; Baker, Johnson & Roberts 1989; McGinnis 1998). Under this hypothesis, these verbs are essentially unaccusative. In relation to the issue that concerns us, the analysis predicts that v_Aux and T share co-indexed person features here (as in (37)), correctly yielding be-insertion.

In Section 6.2, I concluded that we should reject the hypothesis that be is associated with unaccusativity (including constructions from (41) to (43)) and have with transitivity.^[22] While making correct predictions in most cases, this approach fails to account for cases where a reflexive clitic co-occurs with a direct object, a context where we find be. Consider the constructions in (44), where the v-head is not defective: here, v probes the IA in the complement of V and assigns accusative case. Unlike in examples (41) to (43), the IA is not promoted to subject position, and it is clear that the reflexive does not absorb the EA’s θ-role. In this context, the subject must be an agent.

In absence of the reflexive clitic, this environment forms the periphrastic past with have, as shown in (45a) and (46a). The presence of se always triggers be-selection, see (45b) and (46b).^[23] Note also that past participle agreement morphology is set as default, unlike the constructions given in (41), (42), and (43). This is particularly clear in (46b), where lavé shows agreement with neither the subject Dominique nor the object les mains, which are both feminine.^[24]

These data indicate that true transitive verbs select be under certain circumstances. As it stands, this evidence confirms that the nature of v is not the cause of auxiliary selection, which goes against the traditional assumption put forward in the literature. In French, it is not transitivity or lack thereof that ‘decides’: be is found when the IA of a lexical verb surfaces in subject position (as discussed in Section 5 notably), and when the reflexive clitic (hosted by the auxiliary) is bound by the subject. If the proposal drawn so far is correct, v_Aux must have an identical person index with T in (45b) and (46b). I present a formal analysis for this syntactic environment in the following subsection.

7.2 Reflexive transitive verbs

Let us build on the analysis developed throughout this paper to posit a formal account for why be spells out with transitive verbs that have a reflexive clitic. In what follows, I first examine the content of reflexive clitics before addressing their syntax. In anticipation of the discussion that ensues, I take these particular clitics to have an impoverished set of ϕ-features, consisting of a person feature only (Amato 2022, 2023; Heinat 2006; Kayne 1993, 2003; Mendikoetxea 2008; Reuland 2001). These theoretical assumptions are then applied to the syntactic analysis of auxiliary selection developed here.

The content of (reflexive) clitics has been a long-standing subject of debate. In my view, Kayne (1993) convincingly demonstrated that Romance clitics could be divided into two categories: person clitics (e.g., me, te, nous, vous, and as we will see below, se) and determiner clitics (e.g., le, la, les, lui).^[25] While person clitics carry a person feature, they lack gender and number features (there is no feminine form ma/ta, and no plural form mes/tes for object clitics). The claim that person clitics lack number might require some justification, as the traditional view treats me as [1,sg] and opposes it to nous as [1,pl]. Nonetheless, Mendikoetxea (2008) argues that the distinction between these clitics is more comparable to the difference between [1,sg] and [1,pl] on verbal affixes (e.g., mange vs. mangeons), suggesting that these features are more verbal than nominal; assuming that number is not intrinsic to person clitics, an alternative characterisation could treat me as [1-person], te as [2-person], nous as [4-person], and vous as [5-person] (the exact label does not matter, it is used descriptively here). Determiner clitics, on the other hand, clearly encode gender and number (cf. la and les). According Kayne (1993), determiner clitics lack person features, with third person considered a case of ‘non-person’ (a notion that goes back to Benveniste 1966). What sets apart person clitics from determiner clitics, under this hypothesis, is that the former only bear [Person], whereas the latter only bear [Gender, Number].

As for the reflexive se, empirical evidence strongly suggests that it should be included in the person clitic group, in that it also seems to lack gender and number (similar to other person clitics, it never inflects). Notably, consider examples like (47) where the reflexive is co-referent with a feminine, plural DP, yet agreement on the past participle is set on default.^[26]

In Kayne’s (1993, 2003) system, se bears a person feature that is set to [0-person], which is not a complete absence of a person feature, but a distinct specification (it is not [3-person] either, given that this author supports the view that [3-person] is ‘non-person’). [0-person] represents a kind of default, meaning that it is neither [1-person], [2-person], etc., but still involves person-related content (i.e., it is not ‘non-person’). In other words, [0-person] signals that the reflexive clitic does not introduce an independent referential entity with its own person features, as a ‘regular’ pronoun would. Instead, it relies on being bound by an antecedent to acquire its person identity (this is Binding Condition A). Kayne (1993: p. 16) further makes an interesting observation, in claiming that his so-called [0-person] (i.e., the default value of the reflexive) “is less strong than a ‘positively numbered’ person”. In more concrete terms, this can be linked to its need for a binder.^[27] This is what differentiates the reflexive from other pronouns, the latter are not set with a default value and (therefore) are not subject to the same binding condition. I assume, then, that the reflexive is merged with a default person feature and awaits binding to be set for [1-person], [2-person], etc., yielding the reflexives me, te, etc. and acquiring an identity. In my view, we do not need to consider [3-person] to be ‘non-person’. Rather, the reflexive is set to [3-person] if it is the value of its binder (see also D’Alessandro 2007, who argues that se is merged with a [3-person] feature). In sum, reflexive clitics have an impoverished set of ϕ-features, consisting solely of a person feature that is identical (and therefore co-indexed) to that of their binder.^[28] This simple assumption naturally aligns with the analysis of auxiliary selection developed here, where identity matching between two functional heads is key.

Before looking at the derivation of auxiliary selection with transitive reflexives, let us consider the syntax of se.^[29] One line of enquiry assumes the reflexive clitic to be phrasal, a DP, generated either as a complement of the verb (Amato 2021, 2022; Kayne 2003; Sportiche 2014), merged in the specifier of an Applicative head in the case of dative reflexive clitics (Amato 2023), or in the specifier of VoiceP (Alexiadou, Anagnostopoulou & Schäfer 2015; Schäfer 2008). Another approach considers the reflexive clitic to be verbal, either a v-head (Armstrong 2013; Basilico 2010; Cuervo 2003; Folli 2002; Folli and Harley 2005; García-Pardo 2021; Jiménez-Fernández and Tubino 2019), a Voice-head (Labelle 2008; McGinnis 2022). Looking at consumption verbs, Folli and Harley (2005) claim that the reflexive clitic is merged on a v-head (specifically, v_Caus) because it functions as a light verb whose specifier introduces the EA. The Italian reflexive clitic si takes on the role of encoding causation, involving a result state. They illustrate this by contrasting sentences like (48a) and (48b), showing that the addition of si changes the argument structure and the event telicity, thus requiring the presence of a small clause complement. The reflexive clitic, therefore, merges into the v position to introduce causative meaning and enforce the event’s telicity.

Focusing on transitive reflexives that introduce an agent EA in Spanish, Armstrong (2013) also claims that the reflexive clitic is merged on v to account for their unique syntactic and semantic properties. Specifically, this author argues that the clitic spells out a special type of v-head (v_Do), which introduces an agent argument while enforcing additional agent-oriented conventional implicatures, such as willful intent and unaided performance. These implicatures make the agent more involved in the event, giving such constructions a heightened sense of agency. Additionally, the event features on v ensure that the verb must have an accomplishment interpretation. This approach offers a granular explanation to the diversity of reflexive constructions.

The advantage of the line of research entertained by authors like Folli and Harley (2005) and Armstrong (2013) is that it unifies the semantic and syntactic properties of the reflexive clitic within a single locus: v. As I will demonstrate below, this assumption fits in with the auxiliary selection analysis presented here. Recall that be-selection occurs when T and v_Aux have co-indexed person features, and that se bears the same person index as its binder. I will demonstrate that T probes the binder while v_Aux probes se, resulting in both functional heads acquiring identical person indices.

Based on the aforementioned theoretical assumptions, the derivation of auxiliary selection with reflexive transitives is illustrated step by step below for example (49).

Following Folli and Harley (2005) and Armstrong (2013), amongst others, the reflexive clitic is merged on v_Prt, and we have seen above that it comes a default person feature, but no identity (50).^[30] Building on the mechanism outlined throughout this paper, V moves to v_Prt, a head that introduces the EA and probes the IA, copying its ϕ-features.^[31] In this case, however, v_Prt is a probe for gender and number only, since it already bears a default person feature in the form of se.

(50)

At this stage, merging of the EA binds the reflexive, which acquires a matching person value and identity (which is identical to that of the EA). As usual, v_Prt transfers its person feature (via the reflexive this time) to v_Aux (51) – this sharing operation is only possible since v_Prt and v_Aux form a complex head that behaves as a complex probe. This follows the same mechanism outlined for transitives in Section 6, except that the person feature belongs to the reflexive clitic rather than the IA, a distinction that is crucial, as it severs the relationship between the IA and v_Aux; otherwise, have would eventually surface.

(51)

The EA then moves to the specifier of TP to satisfy the EPP and be probed by T, while [se+v_Aux] incorporates to T to yield tense and agreement morphology on the auxiliary (52). This structure further supports the claim that be spells out when the person features of v_Aux and T have the same identity: since the EA values T’s person feature and binds the reflexive, and the reflexive values v_Aux’s person feature, both T and v_Aux receive co-indexed features.

(52)

The advantage of this proposal is that no additional mechanism of auxiliary selection is needed in the case of reflexive transitives. The core structure remains the same whether the verb is unaccusative or transitive, and whether a reflexive clitic is present or not.