2.1 Crosslinguistic properties

The middle reading is a crosslinguistic phenomenon, yet natural languages differ as to the sort of strategy they use in order to encode this construal in their grammar. In fact, middle formation is distant from being a uniform process across languages, or even within the same language; hence some scholars consider the middle as a semantic category, rather than a syntactic one, that can be conveyed via a variety of syntactic constructions in each language (Condoravdi 1989; Lekakou 2005). Ackema & Schoorlemmer (2006) point out that there are languages whose middles contain a reflexive marker, and those which do not; middle constructions can be agentive or non-agentive, personal or impersonal, or be licensed as arguments or adjuncts. This crosslinguistic heterogeneity sparked syntacticians’ interest in this phenomenon throughout the years, which translated into numerous analyses of different nature, including syntactic (Keyser & Roeper 1984; Hale & Keyser 1986; Roberts 1987; Stroik 1992; Schäfer 2008), semantic (O’Gready 1980; Dixon 1982; Chierchia 2003) and lexicalist analyses (Fagan 1992; Ackema & Schoorlemmer 1994; 1995), to name a few.

Generally speaking, these are generic non-episodic predicates conveying inherent properties of a notional object, which has been promoted to be the grammatical subject, while the notional subject has been demoted, albeit its presence as an arbitrary external argument appears to remain semantically.

In (5), the verb’s internal argument becomes the grammatical subject which shows agreement with the verb; moreover, the possibility of reintroducing the verb’s external argument by means of a by-phrase is ruled out. Nonetheless, it is interpreted arbitrarily, which contributes to the structure’s generic reading. This genericity can denote potentiality, i.e. anyone’s capability of exercising a given event on a particular theme: e.g. Anyone can read this article easily (cf. (5a)).

Some scholars (Oosten 1977; Condoravdi 1989; Hulk & Cornips 2000; Ackema & Schoorlemmer 2006) point out the fact that the theme appears to entail some degree of agency, in the sense that it is some of the notional object’s properties that determine the progress of the event represented in the middle. Thus, Condoravdi (1989) argues that the type of adverbs that can appear in these structures must specify the mapping of those events onto time, or else determine the theme’s degree of affectedness by the event over time. In other words, the object measures out the event, and this can be done by calculating how affected the former is by the latter.

Additionally, the generic reading is conveyed by means of imperfective tenses, which can turn an otherwise episodic predicate into a non-episodic one; thus, middle constructions cannot occur in the progressive (6a), or the imperative (6b). Should the verb be conjugated in a perfective tense (6c), the sentence is no longer interpreted as generic, but as episodic, therefore losing the middle reading.

Ackema & Schoorlemmer (2006) (henceforth, A&S) propose a crosslinguistic classification of middle constructions based on whether such structures in the different languages lack an identical episodic counterpart (Type I middles) or have one (Type II middles). Thus, middles in languages like English or Dutch would qualify as Type I middles, whereas Spanish middle-passives (7a), which can receive a passive interpretation (7b), would be considered Type II middles, in A&S’s terms. In fact, these structures’ episodic counterparts can occasionally license by-phrases (7c) just like periphrastic passives do (7d), although this possibility is very limited and contingent on specific registers, such as legalese.

Despite this crucial difference, there are certain properties that remain constant across languages. First of all, it is generally agreed (Levin 1982; Keyser & Roeper 1984; Fellbaum 1985; Condoravdi 1989; Fagan 1992; Ackema & Schoorlemmer 2006) that genericity is at the core of these structures; they portray as genetic an otherwise episodic predicate; the use of imperfective tenses as well as the fact that in many cases the notional object or grammatical subject stands for a type (cf. (5b) above), or as a specific token pertaining to that type, contribute to this type of reading. This is not to say, however, that middle constructions cannot occur with grammatical subjects standing for specific concrete tokens (cf. (5a)); in such cases, the generic reading is preserved, arguably due to the atemporal potentiality these structures convey, i.e. instead of punctual actions, these structures express potential events conditioned by the inherent properties of the grammatical object, as well as the arbitrariness of the implicit external argument in them,⁴ which can be rephrased by one, or people in general (Fagan 1992).

Modification by an adverbial, or in some cases a prepositional phrase, seems to be a common requirement of middle constructions crosslinguistically; in the presence of negation, or a modal auxiliary, this requirement can be waived. The role of this adjuncts, according to Condoravdi (1989), is to presuppose and modify the implicit agent’s active and volitional involvement in the event.

Another common trait of middle constructions across the board is the verb’s internal argument’s promotion to grammatical subject from its complement position. This phenomenon has been considered an instance of A-movement similar to the one found in regular passive sentences (Chomsky 1981; Jaeggli 1986), unaccusative verbs (Burzio 1986) or raising predicates (Chomsky 1981), i.e. the grammatical subject initially merges as an object to the verb and subsequently undergoes movement to subject position, presumably to conform with the EPP; in such analyses, middle constructions are considered unaccusative predicates. Others propose that the internal argument merges as an external argument from the very beginning (Roberts 1987; Ackema & Schoorlemmer 1994); in that case, middles would qualify as unergative structures. For my analysis, I assume the former approach, i.e. that the verb’s internal argument merges inside the VP, as complement to the verb, both in the generic se-passive and in the generic inchoative configurations, and is subsequently probed to preverbal subject position, which I will assume to be SpecTP, if it is available.

Finally, and most importantly for the purpose of this paper, the demotion of the external argument is widespread in middle constructions, which emphasizes the notional object’s intrinsic properties that condition the event. Although most analyses of these constructions concur with the fact that this implicit argument is semantically present, there are different opinions as to whether it is syntactically absent or not. In the following section, I review the main proposals regarding this issue.

2.2 Implicit external arguments in middles

Although most analyses of middle constructions concur with the fact that the verb’s external argument is semantically implicit, there are different opinions as to whether this argument is syntactically absent or not. Thus, while proposals such as van Oosten (1986), Zribi-Hertz (1993) and Ackema & Schoorlemmer (1995) reject the idea that the syntax projects the verb’s external argument in middle constructions, others like Hoekstra & Roberts (1993) propose that Type I middles contain an arbitrary null pronoun (pro^ARB) in the specifier of the verbal phrase that stands for the implicit agent, or Stroik (1992; 1995; 1999), who claims the external argument merges as PRO in adjunct position. In this section, I explain how some of the most salient analyses of middle constructions, including syntactic and lexicalist approaches, account for the formal encoding of this argument.

3.1 Inchoative structures in Spanish

Inchoative constructions denote changes of state, either physical (25a’) or psychological (25b’), undergone by their grammatical subject. These events are predicated primarily by verbs that are subject to the causative alternation (Schäfer 2009, inter alia); when a causative predicate occurs in its reflexive form,¹⁶ it becomes unaccusative and is referred to as ‘anticausative.’ The reflexive clitic’s phi features in these sentences must necessarily agree with those of the grammatical subject; furthermore, the latter is not interpreted as a volitional agent, but as affected by the event, and so is the reflexive clitic, i.e. there is coreference between them. Moreover, the presence of an implicit agent is completely absent in the anticausative construction, and this can be seen in the unacceptability of the insertion of a volitional adverb or prepositional phrase (e.g. adrede; ‘deliberately’); alternatively, these structures license by itself PPs, which is a sign of their lack of agentivity.

Similarly, other predicates expressing changes of location (26a), and inherently reflexive verbs (26b) can also appear in inchoative constructions, although they lack a causative counterpart.¹⁷ As in (25), the grammatical subject in these sentences is also interpreted as affected by the event and agrees in person and number with the reflexive clitic.

Thus, throughout this article I use the term ‘anticausative’ to refer to unaccusative structures containing verbs participating in the causative alternation, as in (25), and ‘inchoative’ as a more general category encompassing both anticausatives and constructions like the ones in (26).

Interestingly, these predicates, although not necessarily, optionally allow for their grammatical subject to be agentive when it is [+animate] and [+human] or at least able to act willfully;¹⁹ in such contexts they are able to license an agentive AdvP or PP.

As I mentioned above, inchoatives, like se-passives, do not project an external argument in the syntax and it is the internal argument that is promoted to grammatical subject; however, in the latter, but not in the former, an implicit agent is interpreted. Another common feature that inchoatives share with se-passives is that in most cases, the unmarked word order seems to be that where the grammatical subject appears post-verbally, as shown in an out-of-the-blue context such as (28):

Nonetheless, unlike reflexive marked passives, the grammatical subject in anticausatives must necessarily be a full DP, not a bare NP; should that be the case, the interpretation automatically switches to passive, as shown by the impossibility of licensing a by itself PP in (29b).

Sánchez López (2002) explains that the events predicated in these constructions initiate spontaneously triggered by some of the intrinsic properties of the affected object, with or without an intervening external cause(r).²² This is due to the fact that these predicates lose the CAUSE argument that they take as a subject in their transitive counterpart. Inchoativity, therefore, is syntactically determined, not semantically.

As for their aspectual properties, Mendikoetxea (1999) argues that these sentences are episodic and not stative, i.e. they denote processes or accomplishments, and are therefore eventive, unlike middle-passives, but they are not effected by an agent, like passives. Therefore, both perfective (30a) and progressive (30b) aspects are possible in inchoative contexts, as well as imperative (30c)²³ and habitual present (30d):

When transitive change-of-state verbs participate in the causative alternation, they are analyzed as containing two subevents: one introducing the external argument, i.e. the causing event, and another introducing the affected internal argument, i.e. the resulting state event. Sánchez López (2002: 81), based on Levin & Rappaport Hovav (1995), proposes the following lexical semantic template for a verb like secar ‘to dry’, in Spanish, which can have a causative representation on the one hand (31a), meaning that this is an event in which x participates and which causes y to become dry; if, on the other hand, secar occurs in an anticausative context, its formal representation would be (31b):

Thus, causative constructions would have both x and y slots filled with an argument, i.e. the event is caused by an external argument; on the other hand, anticausative configurations lack an external causer, which has been morphosyntactically or lexically suppressed. In the case of Spanish and other Romance languages, such as Catalan or Italian, this unaccusativity is encoded by means of the reflexive clitic se, as mentioned above, that blocks the insertion of an external argument in se-passives (32a), inchoatives (32b) and middle-passives (32c).²⁴

The lack of an external argument in subject position in unaccusative se constructions triggers verbal agreement between the verb and its internal argument, which becomes the grammatical subject and bears nominative case. Next, I provide an analysis for inchoative middles.

3.2 The analysis

At this point, it is now time to propose an analysis for inchoative middles. As shown in (33), change-of-state predicates such as quemar (‘to burn’) may occur in what appears to be a middle-passive configuration. Crucially, in addition to the generic se-passive reading associated to these structures (e.g. Anyone can wrinkle linen shirts easily), these contexts display an additional interpretation, i.e. that of a generic spontaneous agentless event of change (e.g. linen shirts have the property of wrinkling easily).

What I propose here is that these two possible interpretations are the consequence of two different underlying representations which happen to have the same realizations at Spell-out, namely, a generic se-passive, on the one hand, and a generic inchoative, on the other. Both configurations, however, predicate intrinsic properties of their notional object, which is one of the criteria used by numerous analyses to determine what middle sentences are. Therefore, I consider these two structures middle constructions, associated under the term inchoative middles.

Cuervo (2003) establishes a distinction between three types of event introducers (v_DO, v_GO, v_BE) which correspond to each of the three simple eventive structures: (i) activities (e.g. to sing); (ii) verbs of change/happening (e.g. to happen), and (iii) states/existentials (e.g. to be happy). Additionally, these introducers can combine with each other, thus creating bi-eventive structures: causatives, which would be the result of an activity event introducer and a state one (v_DO + v_BE), or anticausatives, which arise from pairing an event introducer of change with a stative one (v_GO + v_BE).²⁶ Crucially, according to this proposal, activity events (v_DO) merge as complements to a Voice head introducing an external argument (cf. Marantz 1984; Kratzer 1996), whereas events of change (v_GO) cannot. In light of this classification, I assume that the generic passive reading in (33) is the result of a causative combination of verbalizing heads, whereas the generic inchoative interpretation emerges from an anticausative cluster, whose derivation can be seen below.

In (34), we see a cluster of two event introducers, v_GO and v_BE, creating an anticausative structure; the clitic se spells out the head of change (v_GO), which shares the same phi features as the grammatical subject,²⁷ and renders the sentence unaccusative, therefore without a Voice projection introducing an external argument.²⁸ The complement to the stative v_BE is the verbal root arrugar, and its subject the internal argument (las camisas de lino). The root undergoes local head movement acquiring the flavors of each v (Folli and Harley 2005) and incorporating the reflexive clitic pronoun as it finds its way to Tº, assuming clitics are defective heads à la Roberts (2010).²⁹ Finally, the only DP available, i.e. the internal argument, is the one in charge of satisfying the EPP features in Tº and licensing its interpretable nominative features.

One thing that is still to be explained is the fact that the internal arguments in inchoative and middle-passive contexts cannot be bare NPs, as shown above. I assume that this is due to the presence of a [D] and a [+definite] features in Tº in these structures, which must be checked by a DP sitting in its specifier; thus, a bare NP would not be able to do so. Moreover, I argue that these constituents only raise to preverbal subject position in generic contexts,³⁰ and not in regular inchoative sentences, due to the combination of such features with a generic sentence operator (Gen), also in Tº. The fact that the internal argument in regular inchoatives does not raise to preverbal subject position would be due to the absence of Gen in those contexts.³¹

As for the generic se-passive counterpart of (33), note that the participation of an implicit generic external argument, interpreted as agent or causer, is necessary in this context. Following Schäfer (2008), I propose that the generic se-passive counterpart of inchoative middles, just like generic passives in Greek, projects a Voice head, spelled-out as the non-paradigmatic third person reflexive pronoun se, which passivizes it and denotes an implicit generic notional subject.³² This implies that the reflexive pronoun in this case has a different nature than the one in the generic inchoative configuration, hence the different behavior in terms of agreement with the grammatical subject.³³ Moreover, instead of a first subevent of change (v_GO), which is unable to merge with a Voice head, this configuration has an activity/causative³⁴ event introducer, i.e. v_DO. This combination would generate the derivation in (35).

In (35), the verbal root is generated as the complement to the head of the state event (v_BE), which, at the same time, is the activity event’s (v_DO) complement, and in whose specifier sits the internal argument (la camisa de lino). A Voice projection, which in an active construction would assign accusative case to the internal argument and introduce the external argument, is passivized by means of the reflexive clitic se. Thus, the internal argument, being the only available DP in the derivation, raises to preverbal subject position, checks the EPP, [D] and [+definite] features in Tº and checks nominative case; as in (34), the root undergoes head movement to Tº incorporating the reflexive clitic on its way.

Having established with the structural scaffolding of the two possible realizations of inchoative middles, it is now necessary to account for the fact that these predicates allow the insertion of a non-subcategorized dative argument in them, which can convey three different interpretations, namely: (i) affected by a spontaneous event; (ii) affected by an externally caused event; and (iii) unintentional/accidental causer, as shown in (36) below.

Several studies (Fernández Soriano (1999b); Cuervo (2003); Fernández Soriano & Mendikoetxea (2013)) have analyzed these non-subcategorized phrases as applied arguments i.e. phrases introduced by a specialized applicative head (Applº) which relates them to other arguments or to specific events.³⁵ In her theory of dative arguments in Spanish, Cuervo (2003), based on Pylkkänen (2002), analyzes dative arguments in causative and anticausative constructions as affected applicatives which merged sandwiched between the two subevents (v_DO/GO + v_BE).³⁶ The author argues that the dative DP is applied to the end state of the DP-theme, however it is not directly related to it; therefore, it is not possession of the object that is entailed, but possession of its resulting state. Thus, the applicative head relates its specifier, i.e. the dative DP, with its complement, i.e. the stative event (v_BE), which favors the former’s interpretation as an entity affected by the internal argument’s end state. However, because ApplP has no scope over the first subevent (v_DO/GO) for it merges as its complement, the dative DP cannot be interpreted as having anything to do with the triggering event, either externally or spontaneously caused. Therefore, the notion of affected applicatives can be applied to inchoative middles, obtaining the following derivations for the generic inchoative (37) and the generic passive (38) counterparts.

In (37), we see an inchoative structure comprising a subevent of change (v_GO) and a stative one (v_BE), between which the applicative head introducing an additional argument merges. The applicative head is spelled out as the dative clitic le, which bears an identity relationship with a full dative DP (a María) that appears left-dislocated in the structure. As in (34), the verbal root undergoes head movement to Tº, incorporating all the clitics it finds on its way; moreover, since no Voice projection introduces an external argument, Tº probes the first DP it finds in the derivation, i.e. the theme las camisas de lino, to its specifier. The applicative argument is applied to the theme’s resulting state, favoring the interpretation of being affected by it. Notice the dative argument could be interpreted as possessor of the theme, but not necessarily; in fact, Cuervo (2003) argues that ‘to assume possession is a way of (pragmatically) accounting for the affectedness of the dative as a result of the change of state of the theme object’ (2003: 126). In (38) below, I show the generic passive configuration containing an affected applicative.

The derivation in (38) differs from (37) in that the first subevent is an activity one (v_DO), which is complement to a Voice projection. Voice is passivized and spelled out by the reflexive clitic se, which denotes the participation of an implicit external argument, compositionally interpreted as generic, given the presence of the generic operator Gen. The applicative head, as in (37), merges between the two subevents, therefore still being interpreted as affected by the theme’s resulting state, yet this time, the change of state is externally caused, not spontaneously.

Notice that both in (37) and (38), the dative argument could be interpreted as possessor of the theme, but not necessarily; in fact, Cuervo (2003) argues that ‘to assume possession is a way of (pragmatically) accounting for the affectedness of the dative as a result of the change of state of the theme object’ (2003: 126).

Once two of the possible interpretations of the dative argument in (36) have been accounted for, the unintentional/accidental causer reading is yet to be determined. Cuervo (2003) proposes that the affected applicative head can also merge on top of the change subevent (v_GO). By doing so, the dative argument introduced by this functional projection is applied to the whole change-of-state process, not just to the resulting state; therefore, the dative is interpreted as being unintentionally responsible for triggering the event. The derivation for such context is shown below.

A sensible question to pose now would be if the dative’s unintentional causer interpretation is exclusive of the generic inchoative configuration or whether it is somehow possible in the generic se-passive counterpart in inchoative middles. In order to answer it, let us consider the following passive examples:

One of the structural properties of passive constructions, both reflexive (40a) and periphrastic (40b), is the fact that they allow control into purpose clauses; this phenomenon has been used as an argument for the existence of an implicit notional subject in these configurations (cf. Bhatt & Pancheva (2006), inter alia). Although hardly mentioned in the literature, purpose clauses can also be licensed in middle-passive contexts, as in (41),³⁷ which supports the idea that in these sentences the participation of a generic/universal implicit agent is denoted:

The relevance of (41) stems from its interaction with an applied dative argument as in (42) below:

While in (42a), which lacks a purpose clause, the dative argument conveys the three expected readings pointed out above, i.e. on the one hand, unintentional causer or affected by an anticausative event, both derived from the generic anticausative configuration, and affected by a causative event, from the generic passive derivation, on the other, when the purpose clause is inserted (42b), the only possible interpretation that is available is the third one, i.e. María is interpreted as the possessor of the wood and therefore affected by someone else’s burning it. This suggests that the unintentional/accidental causer interpretation is only possible in generic inchoative contexts; in the presence of an agent of some sort, implicit in this case, the dative argument cannot be interpreted as triggering the event of change. In other words, there seems to be a competition between the implicit notional subject and the dative argument to be responsible for the event or, in structural terms, a competition for the position right on top of the first subevent; nonetheless, an agent, be it implicit or explicit, will always take over, relegating the applicative head to a lower position.³⁹ Below is the derivation for (42b):

As shown above, quemar is a verb that participates in the causative alternation, i.e. it can occur in both causative (Sandra quemó la madera; ‘Sandra burned the wood’) and anticausative contexts. Spanish also has a similar verb which lacks a transitive version, and therefore can only occur in inchoative contexts: arder. Thus, when this verb occurs in a generic context, it is expected to enter the generic inchoative configuration exclusively, not the generic passive one; if this is so, control into a purpose clause should be impossible; this expectation is met:

As we can see in (44a), control into a purpose clause is disallowed with this predicate. Moreover, when an unselected dative DP is inserted in this construction, its only two possible interpretations are (i) unintentional causer⁴⁰ and (ii) affected by the inchoative event, but not (iii) affected by an externally caused event. This is more evidence showing the lack of an implicit agent in this configuration, as expected; the affected applicative head is therefore free to merge on top of the change subevent and entail a slight degree of agency, i.e. be interpreted as the unintentional causer. Crucially, if this is the case, one might wonder whether the dative argument can therefore control into a purpose clause in the absence of an implicit agent, just like this argument would do; this is in fact the case for some speakers, who claim to have the reading in (45).⁴¹

The derivation of (44b), with the affected reading, is shown in (46) below, and the unintentional causer configuration is given in (47), with the applicative head merged on top of the change subevent:

The head of the subevent of change in (46) and (47) is not spelled out as the paradigmatic reflexive clitic pronoun, for the verb arder is a simple unaccusative predicate denoting an internally caused process and is not part of the causative alternation; in other words, because this is a simple unaccusative verb, it does not involve a bieventive structure, but a monoeventive one, i.e. it simply involves one event introducer (v_GO). Nonetheless, the root still acquires its flavor of change when it lands in it as it undergoes local head movement on its way to Tº.

In this section I have provided an analysis for the two possible configurations underlying inchoative predicates in generic contexts, i.e. a generic inchoative and a generic reflexively marked passive, which happen to have the same phonological realization. Moreover, I have shown that while both derivations allow the insertion of a non-subcategorized dative argument, only the generic inchoative one favors the interpretation of this argument as an unintentional causer, for it can merge in the available position right above the subevent of change (v_GO). In the generic passive, this position is already filled with a passivized Voice head favoring the interpretation of an implicit generic agent; therefore, the applied dative argument can only merge below v_GO, being then interpreted as affected by the event.

3.3 A note on Wood & Marantz (2017)

Before moving on to briefly discussing the position preverbal DPs occupy in inchoative middles, let us delve into the question that there appears to be a competition between the implicit notional subject and the dative argument to be responsible for the event in inchoative middles, yet the agent, be it explicit or implicit, seems to always prevail. As I mentioned earlier, in structural terms one could argue that the applicative head and the passivized Voice head both qualify to occupy the position on top of the verbalizing head (v); however, the applicative can only merge there when there is no (active/passive) Voice, i.e. only in the context of v_GO.

Wood & Marantz (2017) propose that argument-introducing functional heads (v, Voice, p and Appl) can all be reduced to one single categorially underspecified functional head i* selecting for a syntactic constituent of category D; this head acquires its specific meaning in function of its surrounding syntactic environment. Thus, an additional argument introduced by i* on top of an inchoative event (v_GO) will necessarily be interpreted as a high applicative, i.e. a non-core argument. If, on the contrary, i* merges on top of a transitive structure (v_DO), the argument it introduces will denote an external argument.

What I claim here is that this is in fact what happens in inchoative middles. These sentences have two possible derivations: a generic inchoative (i.e. inchoative) configuration and a generic se-passive one. Both structures differ in the absence vs. presence of a Voice head passivized by the reflexive clitic se which denotes a notional subject, respectively, yet both allow the insertion of an unselected non-core dative argument. The data in the previous section show that the unintentional causer interpretation of the dative argument is only available in the generic inchoative context, which is indicative of the fact that the position above the verbalizing head is available for i* to merge, i.e. it is not filled by a Voice head, and because this is a spontaneous event of change (v_GO), the underspecified argument-introducing head i* is realized as a high applicative. The fact that the affected reading is only possible in the context of an active or passive causative event suggests that the position atop v_DO is already filled by Voice, forcing i* to merge below it and be realized as an affected applicative. Notice that both applied argument merge in different positions, yet when the derivation becomes linearized they get pronounced in the same position, which gives rise to the syntactic ambiguity discussed here.

If this proposal is on the right track, one would expect to never find an unintentional causer interpretation of a dative when there is an active or passive Voice head; the following data shows that this is in fact the case.

What (48) shows is that whenever a Voice head is present, be it active (48a) or passive, independently of whether it is in a periphrastic (48b) or reflexively marked passive (48c) construction, it will always be the explicit or implicit agent that can be interpreted as the unintentional causer, while the dative argument is merely affected by the theme’s resulting state. However, in an inchoative context (48d), the dative DP can be understood as the accidental causer of the change of state, for there is no other eligible argument in the structure, i.e. there is no Voice head introducing a notional subject, that can bear this role. In sum, the analysis provided here supports Wood & Marantz’s (2017) proposal; Table 1 captures this generalization by showing the elements in the Numeration of each derivation, how they combine, and how they are realized.

In the next section, I conclude the analysis of inchoative middles by proposing that dative DPs in inchoative middles can merge in two different positions: either in the specifier of ApplP, therefore being the outermost DP and being the optimal candidate to be probed by Tº to its specifier, or left-dislocated as a topic outside the sentence domain.

3.4 On the position of the dative DP in inchoative middles

Several studies (Fernández Soriano 1999b; Fernández Soriano & Mendikoetxea 2013) claim that datives in anticausative constructions behave as subjects, share properties with Icelandic quirky subjects, and their position in the derivation prevents the DP theme from exiting the VP. In this section, I suggest two possible positions for preverbal dative DPs when they double the dative clitic pronoun in inchoative middles; I propose that these phrases can either merge as left-dislocated elements outside the sentential domain, or they can occupy the position traditionally associated with Spanish preverbal subjects, in which case I assume they raise from the specifier of ApplP. I base this claim on the behavior inchoative middles show when a subjecthood test proposed by Masullo (1992) is applied to our data, explained below.⁴²

Masullo (1992) elaborates an analysis of dative experiencers and shows that these projections have subject-like properties and occupy the specifier of TP. Among the subjecthood tests that this author provides, one crucially triggers interesting effects with regards to the data under analysis here. Masullo (1992) argues that the fact that the preverbal quantifier DP a nadie (‘to nobody’) is only grammatical if non-dislocated proves that this constituent is sitting in an argumental position.

The relevance of the contrasts in (49) comes from the effects they trigger with respect to inchoative middles, which can be observed below:

The unacceptability of (50a) shows that both the negative dative quantifier and the nominative DP theme are competing for the same position in the derivation, i.e. the subject position, which, again, I assume to be SpecTP. However, when the DP theme remains inside the VP, the sentence is fully grammatical and conveys the desired meaning: Asturian potatoes don’t burn quickly for anybody. Should the negative quantifier be left-dislocated, the interpretation the speaker obtains is that there is an entity called Nadie who causes Asturian potatoes to fry fast. In other words, the unacceptability stems from the negative quantifier losing its scope for being left-dislocated, but this is not to say that that position, i.e. SpecTopicP, cannot be occupied by a left-dislocated non-quantified dative DP, as I have been showing in the trees above (cf. §3.2). Crucially, these data prove that there exists a competition between the dative DP and the DP theme for the subject position in these structures, albeit the theme seems to be at a disadvantage for merging lower in the derivation. Nonetheless, should the theme raise to SpecTP, the dative DP will emerge as a topicalized element.⁴³ On the contrary, if only the dative DP occurs preverbally, I assume it is because it generates in the specifier of the ApplP, inside the sentence, and is probed by Tº to satisfy the EPP; however, since the dative DP bears inherent case, Tº will look for the next caseless DP to check its nominative case feature, i.e. the theme, which remains in situ within the VP-domain.⁴⁴ The derivation can be seen below:⁴⁵

In this section, I have provided evidence supporting the idea that dative arguments in inchoative middles, be it the generic se-passive or the generic inchoative configuration, as in traditional inchoative structures, are introduced by an applicative head which relates a dative argument to a spontaneous event of change (v_GO), and is therefore interpreted as an unintentional causer, or to the internal argument’s resulting state (v_BE) in an inchoative or passivized causative context; in the latter cases an affected reading is then obtained. This dative argument, which is spelled out as a dative clitic, can optionally occur reduplicated by a full preverbal dative DP. I argued that two possible configurations can yield these structures: one in which the dative DP is left-dislocated outside TP, thus favoring the DP theme’s reaching its characteristic preverbal position (presumably SpecTP), and another where it is the dative DP that merges in subject position after being probed from the specifier of ApplP by Tº. In such case, the sentence’s internal argument must remain in its base position inside the VP.