3.1 Restrictions based on the features of the dative argument

As shown in the examples from (15) to (19), the dative clitic immediately preceding stylistic LE can only bear 1sg or 2sg features; if the dative clitic is 3sg, 1pl or 2/3pl, the pattern yields unacceptable results. The relevant distribution is illustrated in simplified form in (21).

A null hypothesis regarding this distribution is that it follows from general syntactic constraints applying on clitic clusters rather than on specific properties of the construction. To begin with, some of the unacceptable patterns in (21) could be accounted for in terms of syntactic haplology, i.e., a ban against adjacent identical elements (van Riemsdijk 2008; Nevins 2012; Neeleman & van de Koot 2017, i.a.).

The unacceptability of the Chilean Spanish clusters le le in (21c) and les le in (21e) could be accounted for by appealing to syntactic haplology. That is, these cases might be deviant simply because they contain sequences of two clitics that are “too similar” to each other, i.e., stylistic LE cannot be adjacent to a third person dative clitic because they are identical in some relevant regard. As noticed by Bonet (1995), among others, the identity criterion at play here cannot be formulated in phonological terms only. In the absence of a more precise definition, the fact that both elements are clitics associated to the realization of a non-core dative argument might suffice to consider them identical.

Suppose, for the sake of argument, that (21c) and (21e) can be ruled out in terms of syntactic haplology. In that case, accounting for the distribution of stylistic LE in (21) reduces to explain the unacceptability of nos le in (21d). This is where the problem becomes particularly challenging, as there seems to be no known constraint allowing to capture both the unacceptability of nos le in (21d) and the acceptability of me le in (21a). To illustrate the complexity of the issue, consider the (weak version of the) Person Case Constraint (PCC), a well known condition on clitic clusters.

The PCC successfully captures the unacceptability of a sequence of clitics of the form nos le in ditransitive constructions, as the cluster involves a first person direct object (i.e., nos) and a third person indirect object (i.e., le).

However, the PCC cannot be the restriction behind the unacceptability of nos le in (21d). If this were the case, the analysis would wrongly predict the ungrammaticality of me le in (21a) as well. That is, a standard ditransitive sentence combining a 1sg direct object (i.e., me) and a 3sg indirect object (i.e., le) into the sequence me le is also expected to violate the PCC, e.g., (24). Thus, if we posit that the PCC accounts for the distribution of stylistic LE, this would fail to capture the acceptability contrast between (21a) and (21d).⁴

Rivero (2008) proposes a further condition governing the functioning of clitic clusters: the Quirky Person Restriction. It establishes that first and second person reflexives cannot combine with third person dative clitics. She illustrates the effects of this constraint with examples involving plural reflexives: as can be seen in (25a), pairing the third person reflexive clitic se and the third person dative clitic le is fully acceptable in Spanish; however, if the reflexive is instead the 1pl form nos, as in (25b), the result is deviant. Given that Rivero’s constraint precludes the formation of a clitic cluster of the form nos le, it could be responsible for the unacceptability of (21d).

However, just as the PCC before, the Quirky Person Restriction does not allow to explain the acceptability contrast between (21a) and (21d). While Rivero’s observations are based on data involving plural reflexives only, the restriction she proposes should apply as well to 1sg and 2sg reflexives. In particular, it correctly predicts that the 1sg reflexive me cannot combine with le, e.g., (26). Since this prediction is incorrect for me le in (21a), it follows that the Quirky Person Restriction is not the principle behind the distribution of stylistic LE.

The fact that neither the PCC nor the Quirky Person Restriction captures the distribution of clitics in (21) does not mean that there is anything wrong with these conditions; rather, it says something about the nature of stylistic LE. Both conditions rule out the combination of certain types of elements with a dative clitic pronoun. The reason they fail to capture the behavior of stylistic LE may simply be that this element, despite appearances, is not a pronoun at all.

As mentioned earlier, stylistic LE lacks the denotational properties of standard dative pronouns, i.e., it does not refer to a participant of the event or to a discourse referent. This can be seen in the complete synonymy between sentences containing stylistic LE and their counterparts exhibiting the standard SE+CL_dat cluster: the presence or absence of LE leaves truth conditions and argument structure unaffected. In this sense, LE seems to function as a purely formal exponent rather than as a pronominal argument.

The semantic contrast between standard dative clitics and stylistic LE has grammatical consequences. For instance, while dative clitics in ditransitive constructions can be correferential with a doubling DP, e.g., (27), stylistic LE systematically disallows this possibility, e.g., (28).

Moreover, while standard dative pronouns inflect for Number, e.g., (29), stylistic LE is morphologically invariable, e.g., (30). This further supports the claim that stylistic LE is not a pronoun.

The interim conclusion seems to be that general conditions on clitic clusters such as the PCC and the Quirky Person Restriction do not account for the distribution of dative arguments with respect to stylistic LE. In particular, these conditions fail to apply because stylistic LE, although it has the shape of a dative pronoun, is arguably not one. The upshot is that the patterns in (21) must be captured by appealing to construction-specific properties and mechanisms that treat LE as something other than a dative pronoun.

3.2 Presence and absence of SE

As mentioned in the introduction, a line of analysis for anticausative SE involves the assumption that this element is responsible for the anticausative interpretation of the predicate. A specific implementation of this intuition is due to Chierchia (2004) and Koontz-Garboden (2009). The core idea of their proposal is that anticausative predicates obtain through a process of reflexivization. That is, they maintain that the relation between a causative verb and its anticausative counterpart is identical to the relation between transitive and reflexive predicates: it is mediated by the semantic contribution of a reflexive element such as SE.

Reflexivization can be understood as a semantic operation that co-identifies two arguments of a predicate. Thus, a reflexive element like SE would take as an argument a relation ℜ between two elements and return the set of pairs in which these elements are identical. Koontz-Garboden (2009: 83) formalizes this idea as in (31).

To exemplify, consider the causative form of the verb abrir ‘open’. As can be seen in (32), it denotes a relation between two arguments x and y, in which some action or property β of y causes x to be open.

According to the theory of Chierchia (2004) and Koontz-Garboden (2009), the semantic contribution of SE equates to co-identifying both arguments. That is, the anticausative marker is an element that takes the externally caused event in (32) and turns it into the internally caused event in (33), in which some relevant property of the theme argument x is interpreted as playing a role in the change of state of x.

As a result, an anticausative sentence like (2b), repeated for convenience in (34), receives a reading akin to “some property of the door caused it to become open”.

This theory elegantly derives both the semantics of anticausative predicates and the observation that reflexive and anticausative morphological markings tend to be the same. In particular, the explanation relies on the semantic import of a constituent appearing in both types of construction.

This proposal encounters a significant problem when applied to examples involving stylistic LE. As discussed, this element surfaces with marked anticausative predicates; this is corroborated by the fact that all relevant examples allow interpreting the dative argument as an unintentional causer, e.g., (35). Despite this, the sentence in (35b) does not exhibit the reflexive element SE that is supposed to be responsible for its anticausative interpretation.⁵ Therefore, it provides straightforward evidence against the thesis that SE is instrumental for the semantics of the construction.⁶

In principle, the theory by Chierchia (2004) and Koontz-Garboden (2009) could deal with the absence of SE in (35b) by assuming that the reflexive element in this sentence is phonologically null. Chierchia (2004: 41–42) argues that in anticausative predicates with no SE-marking “the reflexive operator is lexically incorporated into the meaning of the verb without any morphological reflex”. In principle, this idea accounts for verbs such as hervir ‘boil’ in (1) or mejorar ‘improve’ in (36), which participate in the causative alternation without introducing any morphological markings in the anticausative variant. That is, the anticausative interpretation of (36b) obtains from a null reflexive that has been integrated into the semantics of the verb.⁷

The same type of logic could be applied to account for (35b): perhaps SE does not surface in this example because its anticausative meaning is already incorporated into the verb.

This solution, however, is inadequate for the stylistic LE phenomenon. There are two main reasons for this. First, such an analysis blurs the strong distinction between marked and unmarked anticausative verbs. That is, since every verb admitting stylistic LE allows for the alternations in (35), then null reflexives would have to be optional in order to capture that SE is silent in (35b) but overt in (35)c. This does not fit the behavior of anticausative predicates in Spanish: while some verbs exhibit some marginal optionality regarding SE-marking (see example (42) below), anticausative verbs generally either require SE or they do not. For instance, the verb mejorar ‘improve’, which is an unmarked anticausative, rejects the presence of SE, e.g., (37).

Similarly, a verb like quebrar ‘crack’, which is a marked anticausative, cannot form anticausative predicates without SE, e.g., (38).

Thus, marked and unmarked anticausative verbs belong to mostly well-defined classes. Assuming that examples such as (35b) incorporate null reflexives in the same way as unmarked anticausatives do goes against this observation.

A second argument for rejecting an analysis of (35b) in terms of a null reflexive comes from the behavior of stylistic LE itself. The idea that unmarked anticausatives recruit a silent version of SE relies on the assumption that marked and unmarked anticausatives have a parallel underlying representation allowing them to undergo the same morphosyntactic processes, e.g., reflexivization. The stylistic LE alternation provides evidence against this assumption: only SE-marked verbs like quebrar ‘crack’ license stylistic LE, e.g., (39). Unmarked anticausative verbs such as mejorar ‘improve’ cannot host this element (40). Given this stark contrast, the hypothesis that stylistic LE sentences lacking SE, e.g., (35b) and (39b), are anticausative due to the same process as unmarked anticausative sentences, e.g., (36b) and (39a), becomes untenable: if there were a null SE in both (39b) and (40b), why is only the latter unacceptable?

Further evidence that marked and unmarked anticausatives are different for the purposes of the stylistic LE alternation comes from the peculiar behavior of verbs such as hervir ‘boil’. As already shown in (1), repeated for convenience in (41), this verb typically does not require SE to participate in the causative alternation.

However, hervir ‘boil’ does exhibit some optionality regarding its anticausative marking: as can be seen in (42), it is marginally possible to add SE to it to produce an anticausative predicate.

Both anticausative predicates in (41b) and (42) may combine with affected datives with different results: only the SE-marked variant in (43b) triggers the unintentional causer interpretation.⁸ Unmarked anticausatives combined with affected datives never make this reading available in Spanish; the same effect is attested with (40a). The contrast fits well within the traditional observation that marked anticausatives express externally caused changes-of-state, while unmarked anticausatives express internally driven changes-of-state (e.g., Martin 2023): in principle, only the former should be able to license accidental causers.

Since (43b) combines anticausative SE and an affected dative, it is able to host stylistic LE, e.g., (44). Interestingly enough, the variant that does not include SE in (44a) is interpreted just as if SE was part of its structure, i.e., it is synonymous with (43b) rather than (43a). This, once again, proves that examples such as (44a) cannot be equated with unmarked anticausatives and, therefore, they do not involve null reflexives.

At least two conclusions can be drawn from these data points. First, the fact that stylistic LE appears only with marked anticausative verbs shows that some aspect of their grammatical representation triggers the emergence of this element; this may point to a difference in the syntactic makeup of marked and unmarked anticausative predicates. Second, stylistic LE makes the presence of SE optional: the construction preserves the same interpretation no matter this element is present or not in the sentence. As discussed, this optionality cannot be equated with the distinction between marked and unmarked anticausatives, and therefore posits a straightforward challenge to any theory that hinges on the assumption that SE is an essential component of anticausative semantics.

4.1 Assumptions on the syntax of affected experiencers and anticausative SE

I take it that non-core dative arguments in Spanish are introduced through applicative heads (Cuervo 2003; Pujalte 2012, i.a.). According to this, a dative DP expressing a non-core argument merges in the specifier position of a projection ApplP; the dative clitic associated to this DP is the realization of the functional head Appl⁰. The interpretation of these elements depends on the position of ApplP in the syntactic spine. In particular, dative arguments interpreted as affected entities are high applicatives in the terminology of Pylkkänen (2008), i.e., they result from ApplP dominating the lexical projection of the verb. More generally, I adopt the scheme proposed by Cuervo (2003: 113) as the basic blueprint for predicates with affected datives. To illustrate, take the sentence in (45).

The structure in (46) depicts the analysis of this example according to Cuervo (2003). In her proposal, the applicative head surfacing as the 3pl dative clitic les combines with a vP_be denoting a state (i.e., the door being open), while ApplP is selected by a head v_do introducing a dynamic subevent; the combination of v_do and v_be results in the bi-eventive configuration proper of causative predicates; see Cuervo (2003) for motivation and discussion of these verbal projections.

To discuss the structure of anticausative predicates, take the sentence in (47), which replaces the subject Cosmo in (45) for the anticausative marker SE.

The hypothesis to explore here is that the anticausative marker is devoid of semantic content and contributes nothing to the meaning of this sentence; instead, it merely satisfies a formal requirement of some anticausative predicates, which are required to overtly signal the absence of an external argument. This assimilates the functioning of anticausative markers to that of expletive pronouns, i.e., both are dummy objects whose only purpose is to fill a slot in the grammatical representation. Such an approach to anticausative markers has been advanced in different ways by Schäfer (2008), Pujalte & Saab (2012), Wood (2014), Alexiadou et al. (2015), Saab (2020), among others; for concreteness, I will adapt the theory in Schäfer (2008) for the remainder of the discussion.

Schäfer’s (2008) theory revolves around the properties of the VoiceP projection. As is known, Voice is a functional head that introduces a causer in its specifier position (Kratzer 1996). For instance, for the proper noun Cosmo to be understood as the agent of (45a), it needs to be selected as the specifier of VoiceP as shown in (46b). According to Schäfer (2008) and other authors, anticausative predicates recruit a semantically vacuous instance of Voice, i.e., they contain a version of the head that does not introduce a causer. Such a dummy version of Voice, however, still requires a constituent in its specifier position due to its purely formal subcategorization features. By assumption, a reflexive pronoun is introduced in this position as an expletive-like element; Schäfer (2008) speculates that reflexives are particularly well-suited to fulfill this function as they are inherently non-referential. Thus, according to the assumptions adopted so far, an anticausative predicate like the one in (47) receives the schematic analysis in (48b), with the reflexive SE being inserted in the specifier position of VoiceP.^9,10

Notice that the scheme in (48) also differs from (46) in terms of the higher vP layer in the structure. That is, while (46) includes a head v_do, (48) incorporates v_go in the same position. In Cuervo’s (2003) system, v_go introduces a subevent of change. While appearing typically in unaccusative predicates, v_go may combine with v_be as in (48) to compose inchoative events.

A final assumption concerns the meanings arising from the representation in (48). I take it that the non-core dative argument in this structure can be interpreted either as an affected entity (i.e., with the reading ‘the door opened for the neighbors’), or as an unintentional causer (i.e., with the reading ‘the neighbors unintentionally opened the door’). This clarification is necessary as authors like Cuervo (2003; 2020) and Schäfer (2008) conjecture that these interpretations stem from two distinct syntactic structures. While such a hypothesis could be compatible with the proposal to be developed here, adopting it would increase the complexity of the analysis without gaining any empirical coverage.

Consider the sentences in (49). As can be seen, they are all ambiguous between the two interpretations.

While the verb cortar ‘snap’ in (49) allows the 1sg dative pronoun me to be interpreted as an unintentional causer, this reading can be blocked by simply replacing the DP la correa ‘the strap’ with the DP la luz ‘the light’, as in (50). The resulting interpretation is that the speaker experienced a power outage.

The examples in (50) are not ambiguous. If the relevant interpretations relied on different underlying structures, then the unavailability of one of the readings should follow from the ungrammaticality of the corresponding syntactic representation. However, there is no obvious reason why replacing the lexical content of a DP would trigger ungrammaticality. The absence of the relevant interpretation in (50) rather seems to stem from world knowledge: the unintentional causer reading is unavailable in (50) simply because such a meaning would be implausible, as blackouts are spontaneous events. Thus, what determines the ambiguity of examples such as (49) is not their syntax, but what pragmatics can do with the compositional meaning of a structure such as (48).¹¹ In short, the case for positing a distinct syntactic structure for the unintentional causer reading seems somewhat weak when considering the formal properties of the construction (at least for the Spanish data).

There are also practical reasons to disregard a separate syntactic representation for unintentional causers. On the one hand, the fact that a sentence can receive this type of interpretation is completely orthogonal to the stylistic LE phenomenon: as seen in previous examples, the alternation applies no matter whether this reading is available or not. On the other hand, this assumption would make the analysis unnecessarily more complex. Consider the following. The formal contrast attested in (50) already requires positing three individual analyses capturing each of the alternatives. If the ambiguity of each sentence is also to be explained in grammatical terms, then the pattern would require six analyses. This move would be justified if it allowed us to capture some property of the construction but, as discussed, there seems to be no obvious empirical gain in pursuing this approach.

Structures like the one in (48) will be taken from now on as the base configuration over which the stylistic LE phenomenon applies. Therefore, unmarked anticausative predicates, which cannot host stylistic LE, e.g., (40), are assumed to have a different underlying representation. While I remain agnostic about the details of the analysis of these predicates, it is worth mentioning the characterization proposed for them by Cuervo (2003). According to her, these involve mono-eventive structures lacking a vP_be denoting a state. When an applicative argument is introduced in these predicates, e.g., (40a) and (43a), it occupies a position above the projection denoting the subevent of change vP_go. The following scheme, based on Cuervo (2003: 162), illustrates a potential analysis for the example in (40a).

4.2 Stylistic LE as a product of multiple exponence

Let’s consider again the initial data in (5) and (6), repeated for convenience in (52). As discussed, General Spanish generates a clitic cluster of the form SE+CL_dat in contexts combining marked anticausatives and affected datives, e.g., se me in (52a). Chilean Spanish has two further ways of expressing the same meaning: a cluster CL_dat+LE, e.g., me le in (52b), or a cluster SE+CL_dat+LE, e.g., se me le in (52c).

The system depicted throughout subsection 4.1 allows us to posit a syntactic analysis for (52a) in the same lines as (48). The relevant structure is depicted in (53). As can be seen, the 1sg clitic pronoun me is the realization of the applicative head, while SE is inserted in Spec,VoiceP to satisfy the subcategorization features of Voice⁰.

To account for the patterns in (52b) and (52c), it is necessary first to advance a hypothesis about what LE is in this type of sentences. There are a few clues that might lead to an answer. As already discussed, stylistic LE seems to contribute no new meaning to the sentence. While it resembles a 3sg dative pronoun, it is morphologically invariable and exhibits no referential properties. As seen, it surfaces exclusively next to a 1sg or 2sg dative clitic, forming in all cases the sequences me le or te le. Finally, it appears in predicates argued to involve applicative arguments.

I contend that the clitic sequences me le and te le found in the relevant examples are instances of extended or multiple exponence (Matthews 1974; Harris 2017); that is, they involve the realization of the properties of a single syntactic unit through two distinct morphological exponents. Intuitively, this process can be conceived as the splitting of an abstract element into two surface forms: me and le in some cases, and te and le in others. I propose that the element undergoing the “split” is the applicative head Appl⁰.

A simple way of making sense of this “split” is in terms of the role that can be attributed to each resulting exponent. Take again the sentence in (52a) with its corresponding syntactic structure in (53). In this example, the applicative head serves two functions simultaneously. On one hand, it is a functional element introducing a dative argument to the predicate. On the other, it is the dative argument itself; that is, it functions as a first person pronoun. In General Spanish, both roles collapse into a single exponent: the dative pronoun me. Chilean Spanish, however, displays the option of separating these functions into two dative-related elements: LE and me. The former is a pronoun-like form that merely signals the presence of an applicative head in the structure, while the latter is the argument introduced by the applicative head. These elements surface as the clitic sequence me le.

From a purely descriptive and theory-independent point of view, this amounts to assigning LE the role of an applicative marker: an invariable element whose sole function is to indicate that an adjacent dative pronoun must be interpreted as a (non-core) argument of the predicate. Therefore, rather than stylistic LE, a more general and language-neutral designation for the phenomenon under study in this article could be stylistic applicative.

Within the Distributed Morphology tradition, the type of multiple exponence described above is captured through the Fission operation (Noyer 1992; Halle & Marantz 1993; Halle 1997; Arregi & Nevins 2012, i.a.). Fission is a postsyntactic mechanism that takes a terminal node consisting of a bundle of features and splits it into two parts, each of them comprising features that co-occurred within the original node. The rule in (54) illustrates the mechanism.

Adapting descriptions by Arregi & Nevins (2018), I propose that the applicative head undergoing the morphological split carries features that specify it as being a clitic (i.e., [–strong]), which is either 1sg or 2sg (i.e., [+part], [±author] and [+sing]), and has dative case (i.e., [dative]). Fission distributes the features of this head into two exponents, Cl₁ and Cl₂. Cl₁ receives features that define it as a 1sg or 2sg clitic pronoun (i.e., [–strong], [+part], [±author] and [+sing]). Cl₂, on the other hand, keeps features that basically define it as an “inflectionless” dative clitic ([-strong] and [dative]). Finally, I assume that this process takes place only if Appl⁰ is in the context of inchoative predicates containing both heads v_go and v_be. The rule in (55) provides a concise representation of this process.

Once the morphological mechanism responsible for multiple exponence has been established, we can now develop an analysis of stylistic LE sentences that exhibit clitic clusters of the form SE+CL_dat+LE. Consider again the example in (52c). This sentence contains the anticausative marker SE, which, as discussed in section 4.1, is introduced in the syntax as the specifier of VoiceP. I propose that stylistic LE sentences containing SE are derived from the same underlying syntactic structure as standard anticausative constructions with an affected dative but without LE, e.g., (52a); this accounts for why both patterns are identical in meaning. This means that the grammatical representation to which the Fission operation applies in order to derive (52c) is the one already schematized in (53).

When the features associated to the head Appl⁰ in (53) are sent to the morphological component, Fission splits them into two distinct exponents, resulting in the surface sequence me le. The first exponent spells-out the Person and Number features of the applicative head; the latter is, by assumption, the realization of a dative feature lacking Person and Number. The splitting operation is schematized in (56).

The resulting exponents me and le combine with SE to form the clitic cluster se me le that surfaces in examples such as (52c). The whole clitic cluster can thus be regarded as a specific morphological realization of the VoiceP and ApplP projections in Chilean Spanish.

A final remark on the Fission rule in (55) is in order. The rule is selective in two regards. First, it restricts the elements to which it can apply. As can be seen in (55), its application is limited to heads bearing a specific set of features: [–strong], [+part], [±author], [+sing], and [dative]; as discussed, this feature bundle corresponds to the 1sg and 2sg instances of Appl⁰. This restriction correctly captures the distribution of stylistic LE with respect to the features of the dative argument. As shown earlier in (21), repeated for convenience in (57), stylistic LE is unattested with 3sg dative clitics, e.g. (57c), and with plural datives in general, e.g. (57d) and (57e). This follows straightforwardly if the Fission mechanism fails to apply to applicative heads that either (i) lack a [+part] feature and therefore do not encode discourse participants, or (ii) lack a [+sing] specification.

The second aspect in which the rule in (55) is selective concerns the structural contexts in which Fission can apply. Specifically, the operation requires an inchoative predicate derived from the combination of two subevents, a dynamic subevent of change and a stative subevent of result. Following Cuervo (2003), these subevents are introduced by the verbal heads v_go and v_be, respectively; this configuration is precisely the one invoked in the rule. As a consequence, stylistic LE cannot surface in configurations where the applicative head appears without this eventive structure; this includes unmarked anticausative verbs, which, under Cuervo’s analysis, lack the stative subevent introduced by v_be, e.g., (51). This restriction further implies that any other environment in which stylistic LE occurs must involve an underlying inchoative representation. I return to this point later in section 5.

4.3 An alternative way to mark anticausativity

There remains to offer an analysis of sentences with stylistic LE that do not include the anticausative marker SE, e.g., (52b). This amounts to providing an account for the optionality of this element in pairs such as (52b) and (52c). That is, the proposal must explain both the acceptability of anticausative constructions without SE, and the semantic equivalence between sentences that include the marker and those that do not.

The analysis I advance here builds on the idea that anticausative markers have the role of overtly signaling the absence of an external argument in the underlying syntactic representation (Schäfer 2008; Alexiadou et al. 2015). I contend, however, that this requirement is ultimately morphological in nature and must be satisfied in the surface representation of the sentence. Particularly, while this condition is canonically met through syntactic means, e.g., by introducing an anticausative marker in a certain syntactic slot, I argue that it can also be satisfied by alternative mechanisms applying at PF.

The syntactic dimension of this system has been already depicted in section 4.1. As discussed, Schäfer (2008) and others propose that marked anticausative sentences contain a non-thematic version of Voice⁰ that neither introduces arguments nor contributes to event semantics. Although this instance of Voice⁰ is semantically inert, it carries a [D] feature requiring its specifier position to be filled with a reflexive pronoun (at least in languages such as Spanish or German). This system ensures that the semantically vacuous VoiceP projection receives some sort of formal marking at the surface level: the presence of the reflexive flags the underlying syntactic structure as anticausative.

Following the spirit of the strong minimalist thesis (Chomsky 2001; 2004), I contend that the presence of a [D] feature on non-thematic Voice⁰ is meant to satisfy a PF requirement. At the general level, the requirement is that verbs pertaining to a certain class of anticausatives must be overtly marked; as discussed by Alexiadou et al. (2015), the precise characterization of this class remains unclear, but seems to correlate with the relative non-spontaneity of the event expressed by these verbs or with the proportion of its causative versus anticausative uses. Different languages encode this condition in different ways. For the case of Spanish, I propose that the relevant PF requirement can be formulated as in (58).

Being a PF condition, (58) can in principle be met through morphological mechanisms that apply postsyntactically. This is precisely what happens in contexts involving stylistic applicatives: their multiple exponence pattern provides an extra clitic that allows to fulfill (58). As discussed, the Fission operation in (54) produces two exponents from splitting the features of an applicative head Appl⁰. If Appl⁰ is 1sg, Fission produces the sequence of clitics me le; if Appl⁰ is 2sg, the result is the sequence te le. As can be seen in (59), both clitics that surface next to LE are syncretic with reflexives in Spanish. Since, syncretic elements are indistinguishable for PF purposes, they can satisfy the overt-marking requirement in (58) with no need for an anticausative marker in the narrow syntax.¹²

A need to satisfy (58) in this way, however, only arises when no reflexive pronoun is merged in the specifier of non-thematic Voice⁰; this scenario may obtain if no reflexive element is included in the Numeration (Chomsky 1995). In such cases, the subcategorization [D] feature on the Voice head remains unchecked. Following Preminger (2014), Longenbaugh (2019) and Newman (2024), I adopt the view that unchecked features do not immediately produce ungrammaticality: the interfaces will still attempt to interpret and pronounce the resulting structure, and they will filter it out if they cannot process it. This idea is captured in (60).

These assumptions make it possible to account for stylistic applicative patterns of the form CL_dat+LE that lack the anticausative marker SE. Consider the example in (52b). Since this is an anticausative sentence without an anticausative marker, it must be the case that the specifier position of VoiceP remains empty, and consequently that the [D] feature on non-thematic Voice⁰ remains unchecked. The corresponding syntactic structure is shown in (61).

Given the principle in (60), this representation is sent to the interfaces, where it will be filtered out if it violates interface-specific constraints. On the LF side, no problem should arise: since the non-thematic VoiceP projection is semantically empty, no difference in meaning is expected relative to “complete” structures such as (53), which do contain an element in Spec,VoiceP; the representation should therefore be interpretable at the semantic level. At the PF side, however, a violation of the condition in (58) is predicted to occur in most Spanish varieties, since narrow syntax provides no reflexive element to overtly signal that the underlying structure is anticausative.

This issue is prevented with stylistic applicative constructions in Chilean Spanish. As discussed, the applicative head me in (61) becomes the sequence me le due to the instance of Fission depicted in (56). While the clitic LE in the sequence serves as the phonological realization for Appl⁰, the clitic me is reanalyzed as a reflexive due to syncretism. Since PF takes me to be a reflexive, it satisfies the overt-marking condition in (58), and the resulting sentence is acceptable.

We may think of this process as filling the available slots in a template. The surface representation of the syntactic structure in (61) should contain at least two clitics: (i) one corresponding to the VoiceP projection, which marks the predicate as anticausative and satisfies (58); and (ii) another that spells out the applicative head. In principle, (61) provides no reflexive element to serve as the phonological realization of VoiceP, so the derivation should crash at PF, e.g., (62a). However, if Fission supplies an additional exponent that can be interpreted as the externalization of VoiceP, then the resulting phonological representation will satisfy (58) and converge at PF, e.g., (62b).¹³

The analysis developed in this and the previous subsection allows us to propose a typology of alternative realizations for sentences that combine marked anticausatives with affected datives in Chilean Spanish. On one side of the typology are realizations in which the anticausative marker is introduced in Spec,VoiceP. Within this group, two configurations arise. If the applicative head does not undergo Fission, the structure surfaces as the standard Spanish pattern, with a clitic cluster of the form SE+CL_dat, e.g., (52a). If the applicative head does undergo Fission, stylistic LE appears and produces a cluster of the form SE+CL_dat+LE, e.g., (52c).

On the other side, there are realizations in which no anticausative marker is introduced during the syntactic derivation. In such cases, the only convergent option is to apply Fission on Appl⁰, so that the resulting pattern of multiple exponence satisfies the condition in (58) and, in a sense, rescues the derivation. This produces a clitic cluster of the form CL_dat+LE, e.g., (52b). The scheme in (63) summarizes the classification.