3.1 The Role of Activity behind u-Syncretism

3.1.1 Scenario #1: A-movement and Agree

The syntax of a se reflexive sentence is similar to a transitive sentence, but with a crucial difference: in this type of reflexives, Voice does not assign/value accusative abstract Case. According to Pujalte & Saab (2012), this follows if Voice can enter the syntax with or without unvalued φ-features. If it has φ-features, then it values them through Agree against the internal object, like in simple transitives. If Voice is, instead, fully φ-defective, then there is no accusative valuation. Importantly, my general approach does not depend on this particular implementation. Other theories of Case assignment would do the same job, namely, deactivate the internal argument when there is such an argument present in the syntactic derivation. In fact, in passing, I will suggest other ways to approach this particular aspect of the theory in order to stress that I have no particular commitment with any theory of Case assignment (see also footnote 5).

Now, because of the formal defectiveness of Voice, the internal argument of a se reflexive transitive sentence, which has already received the theme θ-role from V, is active with respect to the probe that se instantiates. Therefore, the internal argument raises to Spec,VoiceP, deletes the EPP feature in se and receives an additional agent θ-role from Voice. In the following tree, I illustrate these aspects of the derivation:

(27)	a.	Ana se criticó.
	b.

In this derivation, A-movement of the active KP to Spec,VoiceP creates a syntactic Agree dependency between this argument and the probe instantiated by the clitic. Thus, the clitic values its φ-features as third person singular. Once T is introduced into the derivation (or C, depending on assumptions about the locus of φ-features), it probes into its complement domain, finds KP, which is still active, and establishes an A-dependency with it. As a result of this new instance of Agree, T values its φ-features and KP receives nominative as its case value. The “visible” effects of these abstract Agree relations are morphological agreement between the subject, T and the clitic, and the case form of the subject (zero in this case). Importantly, this system derives all instances of the so-called “paradigmatic” se as the result of syntactic Agree. All the agreeing variants of the clitic (me, te, nos, etc.) obtain their features in the syntax and their surface form at PF. If Agree fails in the syntax, this, in turn, will result in the so-called “non-paradigmatic” se, i.e., impersonal/passive se. On this approach, there is no need for the basic division that most analyses of Spanish se make between the paradigmatic and non-paradigmatic types. In effect, according to the present theory, there is only one se in the syntax that merges with Voice, deletes its [D]-feature but, given its formal makeup (absence of K), cannot receive a θ-role. The main difference between paradigmatic and non-paradigmatic se relies on whether or not the probe instantiated by the clitic succeeds attracting a full argument. As I have argued here, it does in the case of reflexive, and most cases of paradigmatic se (as we will see below), but it does not in non-paradigmatic se scenarios, to which I turn my attention now.

3.1.2 Scenario #2: Agree failure

Let us assume that in impersonal/passive structures with se, Voice, unlike paradigmatic se, is a probe for accusative. Case Empirically, the contrast between the derivation of an impersonal/passive or a reflexive se sentence stemming from the same verbal root is clear and not subject to particular controversies, with the exception of passive se, which I will discuss below. In other words, there is no doubt that the contrast between (27) and (28) below is at least a contrast in the case form of the internal argument. Anyway, let me illustrate the assumption that Voice values the internal argument as accusative through the following tree (double arrow = Agree between Voice and the internal KP):

(28)	a.	Se criticó a Ana.
	b.

Note that Case valuation by Voice necessarily precedes Voice θ-assignment. Otherwise, the agent θ-role would be assigned to the internal argument even in impersonal se environments. Alternatively, one could conjecture that Case valuation applies before than θ-assignment because it depends on a functional head below Voice (e.g., αP, see López 2012).⁶ At any rate, the sole crucial difference between this tree and the tree in (27b) is in the formal content of Voice, which in the impersonal scenario contains a set of unvalued φ-features. Voice itself then establishes an A-dependency with the internal argument and, as a result, this argument is deactivated. The direct consequence of this is an Agree failure between se and the internal argument (see Preminger 2014 for extensive discussion and a theory of Agree failures). By the same reasoning, T cannot value its uninterpretable features either. For impersonal se, this generalized failure is resolved at PF by default agreement and PF deletion of the EPP feature in D. For passive se, a particular rule of postsyntactic agreement relates T and the object (see below). Therefore, the non-paradigmatic nature of impersonal and passive se is just the surface manifestation of this Agree failure and not the result of any intrinsic property of the clitic.

Now, because of this failure and the functional defectiveness of se, Voice does not discharge its θ-role to any KP. In LF terms, this means that the agent θ-role is realized in the Voice head. As we will see in subsection 4.2, this syntactic scenario feeds combination between se and the Voice head under regular Functional Application. I will postpone detailed discussion of this aspect of the theory to section 4.

At first glance, the theory I am developing seems to make an incorrect prediction with respect to passive se constructions, where the internal argument agrees with the verb (cf. (2) above):

1. (29)

1. Se
2. SE
1. cerraron
2. closed.3PL
1. las
2. the
1. puertas
2. doors
1. para
2. for
1. bloquear
2. block.INF
1. la
2. the
1. salida.
2. exit
1. ‘The doors were closed in order to block the exit.’

However, as shown by Pujalte (2012; 2020), Pujalte & Saab (2014), Saab (2014) and Ormazabal & Romero (2020), this is a kind of nominative illusion. Concretely, passive and impersonal se do not differ regarding their abstract Case structure; both instantiate the same abstract syntactic structure deployed in (28a). Therefore, in passive se configurations there is no nominative Case valuation. This can be corroborated in contrasts like the following ones, in which passive se does not admit nominative pronouns or proper names in subject position:⁷

1. (30)
1. a.

1.   Se
2.   SE
1. encontraron
2. found.3PL
1. cadáveres.
2. bodies
1.    ‘Bodies were found.’

1.  
1. b.

1. *Se
2.   SE
1. encontró
2. found.3SG
1. Juan/él.
2. Juan/he
1.    INTENDED: ‘He was found.’

1.  
1. c.

1. *Me
2.   CL.1SG.ACC
1. encontré
2. found.1SG
1. yo.
2. I
1.    INTENDED: ‘I was found.’
2.    (b-c OK as reflexives; see Pujalte & Saab 2012)

Accusative pronominalization or differential object marking must be used here. This always results in an impersonal se sentence:

1. (31)
1. a.

1. Se
2. SE
1. los
2. CL.MASC.3PL.ACC
1. encontró.
2. found.3SG
1. ‘They were found.’

1.  
1. b.

1. Se
2. SE
1. me
2. CL.1SG.ACC
1. encontró.
2. found.3SG
1. ‘I was found.’

1.  
1. c.

1. Se
2. SE
1. encontró
2. found.3SG
1. a
2. DOM
1. Juan.
2. Juan
1. ‘Juan was found.’

For some dialects, the right generalization seems to be that only those objects that are not explicitly marked as accusative show subject-verb agreement effects. In Pujalte (2012) and Pujalte & Saab (2014), this agreement difference between passives and impersonals is considered as purely morphophonological. If this is correct, we have to dissociate morphological case from morphological agreement. We refer the reader to those works and to Ormazabal & Romero (2020) and Pujalte (2020) for recent related proposals.

In summary, I have illustrated the role that the Activity Condition plays in the derivation of reflexives and impersonals/passives. In the next subsection, I turn my attention to the question whether other standard restrictions on A-movement play also a role in the derivation of se reflexives in Spanish. I will show that this is indeed the case and that by assuming that se reflexivization is an instance of A-movement, we can provide a straightforward account for certain intriguing reflexivization patterns in ditransitives.

3.2 Reflexivization in ditransitives: Evidence for A-movement

A crucial ingredient of the present theory is the view of se as a probe for A-movement. Regular diagnostics for A-movement (absence of WCO effects, for instance) cannot be tested in reflexives. Yet, there is empirical evidence coming from a contrast observed in Kaminszczik & Saab (2016; 2017) involving se reflexivization in ditransitive sentences. It is well-known that Spanish has a dative alternation in which the goal KP can surface as a PP headed by a ‘to’ or as a-marked KP doubled by a dative clitic (see Masullo 2003; Demonte 1995; Cuervo 2003; see also Pujalte 2012 for another approach):

The example (32a) illustrates the prepositional variant of the alternation whereas (32b) illustrates what I will call without any theoretical commitment “the double object construction”. Following main insights in the literature, I assume that the dative alternation reverses the c-commanding relations between the two internal arguments:⁸

(33)	Prepositional construction

(34)	Double object construction

Assuming this analysis of the dative alternation, the following prediction arises with respect to reflexivization of the theme argument in both variants: only the theme KP in the prepositional variant can move attracted by se. This prediction is correct. In the following examples, reflexivization of the theme KP is licensed only in the prepositional construction. In the double object construction, the result of reflexivizing the theme KP either is ungrammatical or gives rise to an idiomatic reading, which I indicate with parentheses below. In the latter case, se is not reflexive, but inherent/diacritic (more on this below):

Standard restrictions on A-movement account for why this contrast exists. Concretely, only in the prepositional construction the theme KP is local in the favored sense, as shown in (39).

In the double object construction, the goal KP is closer to the Voice head than the theme KP, which, as a consequence, cannot be attracted by the relevant probe that the clitic instantiates. Therefore, we correctly rule out any attempt to A-move the theme KP to Spec,VoiceP in this syntactic configuration. Put differently, in the double object construction, only the goal KP can be reflexivized.

The following example, in which the goal A-moves to Spec,VoiceP, shows that this argument can indeed be reflexivized:

Three comments are in order. First, we can assume here that dative Case is structural (at least in ditransitives), but, unlike nominative and accusative, is not valued by Agree with a given functional head, but through a PF mechanism that inserts the preposition a to an argument that has not valued its Case feature in the syntax. Pujalte (2012) motivates this operation by properties of the inheritance mechanism and the particular distribution of dative arguments in the syntax. I refer the reader to her work for details. Alternatively, dative Case is valued via Agree with a probe above VoiceP. Under both alternatives, the goal KP is active to get a second θ-role from Voice. At any rate, it is important to stress that these are auxiliary assumptions. In principle, the present theory is perfectly compatible with other Case theories offered in the literature. A prominent family of theories are the competition-based ones (among others, Marantz 1991; Preminger 2014; Baker 2015). According to such theories, nominals obtain their abstract Case value from competition among them in a certain structural domain. We can, then, adapt the present theory to such views by exploiting a particular feature of my analysis. In effect, recall that I have assumed that clitics are hybrid regarding their phrasal status. Following ideas by Chomsky (1995) and others, I have implemented this assumption by stipulating that clitics are minimal and maximal:

(42)	Dmin/max

Now, let us assume that for reflexives the Case calculus exploits the feature [minimal], but that for impersonals, the feature [maximal]. If this is correct, Case determination in (28b) should be replaced by the following alternative analysis (double arrow = argument competition):

(43)	a.	Se criticó a Ana.
	b.

This is an interesting move, with clear consequences for the issues I am exploring here. If this option is plausible, then the story would go as follows. In reflexives, at least as far as Case is concerned, D is [minimal] and, consequently, invisible for Case competition.⁹ For this reason, in reflexives the internal argument is active and can be attracted by se. In impersonals, the clitic, being maximal, counts for Case competition. Roughly speaking, once the clitic merges with Voice, it competes for Case with the internal arguments. Following roughly Baker (2015) and others, since the clitic c-commands the internal argument, this argument receives accusative with the consequence of becoming inactive for A-movement and further θ-assignment. Note now that on this approach the impossibility of reflexivizing the theme argument in double object configurations follows from Activity and not from Locality. Merging the goal argument above the theme results in accusative assignment to the theme argument and in its consequent inactivation for establishing further A-dependencies. Either way, we correctly rule out all the ungrammatical sentences we are discussing in this section and correctly rule in the grammatical ones.

In principle, it is hard to see which theory is better, since this depends to a great extent on theories concerning the morphological case/abstract Case connection and, as I said, this is something that cannot be resolved here (see footnote 5). It seems that the competition-based approach handles in a better way the fact that we can have pairs like the following ones, where the goal KP seems to be active or inactive depending on whether the sentence is reflexive or impersonal:

The sentence in (44a) was already discussed above, when I explained why only the goal argument can be reflexivized in the double object construction. The sentence in (44b) has an impersonal reading, which is derived from the same underlying double object configuration. It is easy to see how a competition-based theory of Case would derive each sentence: either se is taken as minimal and we obtain the reflexive sentence or is taken as maximal and we obtain, instead, the impersonal sentence. Under the Agree theory of Case, it is not evident how the impersonal sentence is ruled in. In principle, if the goal is active when se is introduced, then it should be attracted to Spec,VoiceP obligatorily ruling out sentences like (44b). In order to account for pairs like these, the Agree based theory would perhaps require additional assumptions regarding the timing of Case valuation, in general, and the nature of the dative, in particular. But since there is no broad consensus regarding the nature of Spanish datives, we cannot resolve the issue here. Importantly, it seems that the theory should acknowledge the non-uniform status of datives (see Pujalte 2012 for a concrete non-uniform proposal). This is clear if we take into account that some datives do not intervene in A-movement. As I have already observed, sentences like (35b) are grammatical if interpreted not as reflexive sentences but as containing a type of inherent se clitic. As we will see in section 5.1, inherent se sentences also involves A-movement. If this is the case, then we are led to conclude that either the dative has inherent case or it is higher in the structure, both plausible solutions from what we can deduce from the vast literature on the issue.

Second, assuming that in double object constructions V enters the syntax with two θ-roles to assign, we should wonder why both θ-roles are not directly assigned to the object KP once V and this KP are merged, i.e., before the introduction of the indirect object:

If this happens, the indirect object would not receive a θ-role and, consequently, the resulting sentence would violate the Θ-Criterion. It seems then that the VP projection must be completed by External Merge of the indirect object before θ-assignment of the goal θ-role:

Yet, this is not necessarily so. In principle, one could let (45) to take place in the syntax and to filter the result at LF. Alternatively, the step in (45) is blocked in the syntax by principles that regulate the timing of External Merge and θ-assignment. This second strategy is discussed in detail in Kaminszczik & Saab (2016). Any decision in this respect would crucially depend on the assumption that a given thematic head can have more than one θ-role and, obviously, on assumptions about the proper analysis of double object constructions. Space limitations prevent further inquiry into this technical issue.

Third, an anonymous reviewer wonders to what extent the evidence discussed here in favor of the thesis that se is a trigger of A-movement is dependent on the analysis of the dative alternation briefly introduced in this subsection, according to which the double object variant and the prepositional one are partially derived by reversing the c-commanding relations between the goal and theme arguments in each variant. This is an important question, especially, in view of the fact that standard tests to detect c-command relations between the relevant nominal phrases do not give rise to robust conclusions. Pujalte (2008; 2009) has shown that not all speakers react as predicted by the analysis of the dative alternation presented here. In particular, for many speakers presence or absence of the clitic does not produce any particular reaction regarding c-command relations. Exactly the same point is made in a recent paper by Pineda (2020). Both authors conclude then that with or without the clitic, the goal argument always c-commands the theme argument. For Pineda in particular, the clitic is the realization of a low applicative head introducing both internal arguments (see also Cuervo 2003). I find this hypothesis implausible. There is robust evidence that in the clitic doubling variant, the goal argument crosses the subject via A-movement. For instance, WCO effects are blocked whenever there is dative doubling, a clear indication that the first step of movement of the goal argument to Spec,CP is of the A-type:

Put differently, what the clitic clearly indicates, regardless of the c-commanding issue, is that the goal argument has to be active in the clitic doubling variant, otherwise A-movement of the indirect object should be blocked by Activity. It is reasonable to suppose that in the non-doubled variant, either the goal is a PP or has inherent Case. Under both analyses, presence of WCO whenever the clitic is absent is correctly predicted for a sentence like (47). Crucially, under both analyses, we also deduce why the goal argument does not intervene when the clitic is absent, even if we accept with Pujalte and Pineda that in the non-doubled variant, the goal argument c-commands the theme argument. If the inherent Case approach is favored, then, of course, we need to make the further assumption that there is no defective intervention.¹⁰

In sum, in this section, I have argued that standard conditions on A-movement, essentially, Activity and Minimality, are attested in the derivation of impersonal and reflexive sentences containing the clitic se. In the next section, I focus on the semantic aspect of my theory in order to show that the syntactic derivations proposed so far give rise to two different inputs to LF.

4.1 Basic assumptions

I propose that, at LF, θ-roles are semantically realized as functions from entities to event predicates (i.e., objects denoting in <e,<s,t>>), like V and Voice in Kratzer (1996). At least for the basic cases, these are the only semantic objects of this type, given that I assume a radical nonprojectionist view on thematic relations, according to which verbal predicates only denote event predicates (see Pietroski 2005 and Williams 2005, among others). A DP then cannot be combined directly with the verbal predicate for semantic reasons. For simple e-denoting DPs, direct combination with V_<s,t> would produce a noninterpretable object at LF:¹¹

In order to resolve this issue, let us first assume the following three axioms for semantic composition of binary nodes:

Second, I assume that θ-roles have a specified denotation at LF; concretely, they denote functions from entities to event predicates:

This removes the predicative import from Voice. Generalizing proposals in Myler (2014), Alexiadou et al. (2015), Wood (2015), Wood & Marantz (2017) and Schäfer (2017), I adopt the idea that Voice denotes the following identity function (maybe partial, if we want to introduce aspectual conditions, for instance):

The logic of the argument leads us to conceive of θ-roles as elements that are assigned/discharged on nominals in order to produce a legitimate predicative structure, according to which K, once it gets at least one θ-role, can take its complement DP as argument. As a concrete implementation, I have already assumed that DPs project a K head that is the receptor of a given θ-role. Verbal predicates are, in the general case, pure event predicates and not internal argument introducers, an idea in accordance with Pietroski (2005) and Williams (2005), among others. Put differently, on this view, K heads are derived argument introducers. In this way, we motivate the need for an argument to have Case through a condition at play at the LF interface. In the abstract, K must receive a θ-role for interpretability considerations at LF and for producing a semantically well-formed combination:¹²

VP and KP cannot be interpreted by FA but by PM:

Again, the idea is not a novelty; it is at the heart of Pietroski’s (2005) event semantics, according to which Merge is semantically equivalent to conjunction and not to FA. Here, I will not go as far as Pietroski, because I assume that Merge can indeed have as semantic correlates both FA and PM. At any rate, I do agree that this approach brings much more advantages than problems. In particular, it avoids stipulating conjunction as part of the predicate denotations and deducing it directly from PM. In what follows, I show how the theory proposed in section 3 and the assumptions about the semantic architecture just discussed offer an alternative to standard event semantics that it is extensionally equivalent to it in a number of cases, but superior in others. But before entering into detailed analyses, a clarification is in order.

As conceived here, structural K starts its syntactic life just as an empty semantic node. There are two initial points to make with respect to this. First, if K does not receive a θ-role, then either K is treated as an empty node at LF or it denotes an identity function. The move is well-known in contemporary formal semantics. This strategy avoids some of the criticisms that the Visibility Condition has received in the past (see Lasnik 2008 for an important overview). Expletives are a case at point.¹³ If expletives are K projections –perhaps with an empty semantic D head encoding, say, person features or other relevant category or inflectional properties– merged in Spec,TP, then K would not receive any θ-role, as desired, and would have no LF import at all:

Second, other circumstances are also worth mentioning. It could be the case, for instance, that KP occurs in syntactic-semantic configurations in which it does not receive any θ-role because it is the argument of an <e,t> predicate. If its DP complement denotes, for instance, in e then, K either is empty at LF or denotes an identity function:

Of course, a proper analysis of copular or expletive sentences largely exceeds the limits of this study. These brief comments only aim to show what I believe is the right way of conceiving of the different semantic realizations of K, namely, as allosemes in the sense of Marantz (2013) and Wood & Marantz (2017). Put differently, K semantic realization is contextually determined by the syntactic environment in which it is allowed to occur. Of course, allosemy should be properly restricted and I will do it in subsection 5.1, where I exploit this aspect of the theory in order to fill a gap in the syntax and semantics of se, namely, inherent se constructions.

4.2 SE reflexives

Let us see now how the derivation of a reflexive sentence proceeds at LF (cf. (4)).

Recall from section 3 the syntax proposed for a reflexive sentence:

(60)	a.	Ana se criticó.
	b.

Note that there are only two important differences when compared with a regular transitive sentence: (i) the Agree dependency between the clitic and the internal argument, and (ii) the fact that the external θ-role is assigned to the active theme argument. As for the LF implementation of these two aspects, I propose that two θ-roles on the same head amounts to some sort of θ-bundling, which is read off as semantic conjunction at LF (see Reinhart & Siloni 2005). Essentially, this gives us the following semantic realization of any K with two θ-roles:

Regarding the point raised in (i), the Agree dependency between the internal argument and the clitic is translated at LF as index identity, which amounts to saying that the clitic receives the same referential index as the trace of the displaced constituent. Focusing only on the relevant portions of the LF tree for (60b), consider the following LF representation:

(62)	a.	Ana se criticó.
	b.

Since now the index the clitic has is a sister of Voice’, predicate abstraction of Voice’ is triggered:

As advanced in section 2, unlike the implementation in Heim & Kratzer, I propose that A-movement does not leave an index as a sister of Voice’. The clitic itself is the LF realization of such an index as a result of syntactic Agree. More concretely, I am assuming here that (at least some) probes are pure λ-abstractors at LF (or indexes in the original terms in Heim & Kratzer 1998). This view of Spanish se as an abstractor contrasts with the more familiar idea of reflexive se as a bound variable (Embick 2004; Doron & Hovav 2007; Schäfer 2008; 2017; Alexiadou et al. 2015, among others). Truth-conditionally, both approaches are, of course, indistinguishable, but I think that they can still be empirically distinguished on syntactic grounds, as I will suggest in section 6, where I conjecture that the bound variable analysis is correct at least for Italian. Conceptually, the hypothesis of se as a mere abstractor makes se more inactive semantically, a fact that seems to be in consonance with the view of a subset of clitics as purely formal material (say, agreement markers). This puts reflexive se in Spanish (not in Italian, as we will see) on a par with inherent se, which is obviously semantically inert. The thesis is attractive because allows for a conjecture regarding variation in the clitic systems of Romance, namely, the idea that pronominal clitics, which are born as variables, can become probes in the syntax and abstractors at LF because of well-known grammaticalization processes. I postpone the discussion on this issue to section 6.

Coming back to the derivation of our se reflexive sentence, with Heim & Kratzer (1998), I am also assuming that variables can be of any semantic type, so that we can stipulate assignment functions for any semantic type, as well. This is explicitly stated by Heim & Kratzer as follows:

According to my analysis, the variable left by A-movement corresponds to an event predicate variable, i.e., it denotes in <s,t>. The KP Ana, which has the denotation in (61) for the K node, serves as the argument for the predicate opened by index assignment (i.e., the LF realization of se after Agree). This gives us the following denotation for VoiceP:

After the relevant λ-conversions, we end up with a standard event semantics for the entire sentences:

Before closing this subsection, I would like to consider some arguments that have been provided in the literature against one of the aspects of my analysis, namely, its “unaccusative part”. In effect, my analysis shares with an important part of the literature on the topic the idea that in se-reflexivization the KP that gets two θ-roles does not originate as an external argument. The idea comes originally from Marantz (1984) and has been dubbed as the “unaccusative hypothesis” for reflexives, a misleading term, in my view. As already discussed in Embick (2004), to claim that the subject of a reflexive is not the external argument (i.e., a DP in Spec,VoiceP) does not imply that reflexives are unaccusatives. They are not in many respects. The basic functional structure of a reflexive is identical to the functional structure of, say, a transitive sentence (or ditransitive, of course) and not identical to that of an unaccusative sentence, which arguably has as a minimum a different flavor of Voice (see Folli & Harley 2005). Besides this, since its original formulation, arguments have been adduced in favor or against the alleged unaccusative nature of reflexives. Reinhart & Siloni (2005) offer two well-known arguments against the unaccusative nature of reflexives in French connected to the (im)possibility of impersonal constructions with inverted subjects, and to the (im)possibility of en-placement from these inverted subjects. Whereas unaccusative sentences allow for both constructions, reflexive sentences reject them. This kind of argument has been shown as inconclusive by Rooryck & Wyngaerd (2011) and by Sportiche (2014). I refer the reader to those works for detailed discussion. Sportiche, however, claims that, while Reinhart & Siloni’s arguments does not show what has to be shown, there is still another argument that makes the unaccusative hypothesis untenable. This involves association with focus in reflexives and middles. The basic fact is related to the ambiguity of the sentence below:

This sentence has a sloppy reading, according to which Juan is the only one that has the reflexive property, and a strict reading, according to which Juan is the only one that considers Juan intelligent:

The following denials allow to disambiguate the sentence in one way or another:

Crucially, denial of the theme argument is impossible:

This contrasts with middles, which, according to Sportiche, have a clear unaccusative syntax:

Sportiche’s reasoning goes as follows. If reflexives were unaccusative, we would expect an additional reading, according to which association with focus would only affect the theme argument, as in middles, i.e., the reading Juan considers only Juan intelligent should be available. But as the infelicity of (70) indicates, this is not borne out. I think that this reasoning is misleading, since the focus structure in reflexives and middles is clearly different. From the focus marking on the subject in (67), we can derive two different questions under discussion (QuD, see Roberts 2012), namely, Who considers himself intelligent? for the sloppy reading, or Who considers Juan intelligent? for the strict one. Both (67) and the denials in (69) are congruent with one or another QuD, but (70) is not, because in this case the alternatives are computed over the theme object with independence of the agentive subject. Therefore, this denial is not congruent with the focus marking of the original assertion in (67). The sentence in (70) is congruent with a radically different QuD, namely, Who does Juan consider intelligent? In turn, in the middle pattern in (71), both denials are congruent with the QuD that the original assertion tries to answer, i.e., What is eaten with the finger in India? Certainly, the sentence in (71c) seems to presuppose another QuD, namely, What does Indian people eat with the fingers? However, this QuD and the original one entail each other, so the discourse is perfectly congruent. Of course, a full exploration of the interesting connections between focus and reflexivization would take me too far from the original goals of this study, but I think that these considerations suffice to show that there is no expectation that middles and reflexives behave in the same way when it comes to evaluating their behavior under focus.

4.3 Impersonal SE

Let us move on now to the syntax and semantics of impersonal se constructions, whose basic example is repeated below (cf. (3)):

Recall our syntax for passive/impersonal se from subsection 3.1.2:¹⁴

(73)	a.	Se criticó a Ana.
	b.

The more obvious LF implementation for a passive/impersonal se derivation in the present framework is as follows. Since the agent θ-role was not assigned to any KP in the syntax, it is directly realized on Voice in the same way as in Kratzer’s (1996) original proposal, i.e., as a function from entities to event predicates:

Now, I propose that given the clitic se did not form any A-dependency in the syntax, it is semantically realized as a free variable without any referential index; put differently, as a pure indefinite in the terms originally proposed by Heim (1982). This idea of assimilating se to indefinites in Heim’s sense was already proposed by Chierchia (2004) (see also Mendikoetxea 2008 for Spanish). On my account, this is derived from a particular syntactic derivation, one in which Agree failed. If we follow Heim, the variable we obtain at LF ends up existentially closed, unless some operator in the environment unselectively binds se. As is well-known, at least since Cinque (1988), impersonal se has a quasi-universal and a quasi-generic version. Chierchia noticed that the quasi-generic reading also supports E-type readings, a fact that provides direct evidence for the variable nature of impersonal/passive se.

With Chierchia, I then contend that impersonal/passive se is an indefinite that can be unselectively bound in certain contexts. For the basic episodic, existential cases, the derivation is straightforward: se, as an entity variable, combines with Voice through functional application:

But recall that this semantic result is a consequence of a defective probe in the syntax, one that cannot value any of its φ-features. This makes impersonal/passive se in Spanish syntactically different from other indefinites. For instance, as noticed by Saab (2014), episodic impersonal se constructions do not tolerate secondary modification, reflexivization or pronominal binding:

Replacing se by uno ‘one’ renders all these sentences grammatical. This shows that se has no interpretable and valued φ-features. MacDonald (2017a), however, argues that inalienable possession constructions in impersonal se sentences empirically justify the presence of an arbitrary pro in the syntax. The following example is adapted from MacDonald (2017a):

In order to account for both (77) and (78), Ormazabal & Romero (2019) propose that the subject of an impersonal se sentence is a minimally specified syntactic category (see also Rivero 2001). Assuming that this subject lacks inflectional features like gender or person directly accounts for all the cases in (77), given that syntactic binding requires inflectional matching. At the same time, they also account for (78), because the implicit possessor in the DP la mano does not have person features. I am assuming here that the description provided by Ormazabal & Romero (2019) of impersonal/passive se is equivalent to my defective probe. Yet, a crucial difference between their analysis and mine is that they seem to accept the basic division between paradigmatic and non-paradigmatic se (i.e., between impersonal/passive se and the rest). In my view, this type of lexical division loses the generalization that the kind of syncretism we are exploring here is systematic, not accidental. On my analysis, instead, there is only one se that participates of, at least, two different syntactic derivations, one in which there is an Agree relation between the clitic and an active argument, and another one in which there is not. Once this minimal difference is accepted, the rest of the differences between, say, impersonals and reflexives fall in place. An Agree relation in the syntax feeds the interpretation of the clitic as an abstractor at LF, whereas whenever Agree fails, no A-chain can be formed in the syntax and the clitic remains as a LF variable without any index (i.e., as an indefinite in Heim’s terms).

Note now that generic impersonals behave differently with respect to some of the tests that detect syntactic activity. For instance, secondary predication is licensed in generic inpersonal se sentences:

This indicates that generic se sentences require a different syntactic analysis, one that perhaps includes a generic operator in the syntax. This is exactly what we expect if impersonal se is seen as a variable that can be unselectively bound.

In sum, in these two sections, I have provided a detailed syntactic and semantic derivation of a set of argument alternations involving the clitic se. Taking this set as a case study has been proved useful as a way of illustrating the research agenda of this study, namely, providing an explicit theory of Case/θ interactions in syntax and semantics with the aim of making sense of a particular pattern of systematic u-syncretism in Spanish. Importantly, I have brought new theoretical considerations for a positive reconsideration of the Visibility Condition, according to which the role of Case in natural languages is producing interpretable objects at LF.

5.1 Inherent SE

An important consequence of the model developed here is its structural flexibility. In other words, in principle there is no ban for an agent θ-role to be derived within the domain of VoiceP. Among other important consequences, this implies abandoning principles such as UTAH (Baker 1988) that impose strict correlations between θ-roles and syntactic positions. This view has also important empirical consequences in the domain of Spanish se constructions. As is well known, there is subset of quasi-reflexive sentences that seems to resist any principled account. I am referring to inherent se constructions. Here is a list of pronominal verbs that combine with paradigmatic se, taken from Di Tullio (2005):

As noted in the introduction, the two main properties of these predicates are that they cannot occur without the pronominal element, and that they reject transitive uses:¹⁵

In the previous section, I made explicit the hypothesis that K is subject to allosemy, i.e., to syntactically conditioned polysemy. There are, of course, contexts in which a derivation crashes just because the syntax does not provide a suitable output for LF interpretation. Consider the following situation. The category V selects a KP, but does not have any θ-role to assign. Recall that being specified for a [D]-feature is a precondition for being a θ-role assigner. The reverse does not hold: having a [D]-feature does not force the bearer of such a feature to be a thematic assigner. This is obvious for functional heads like, say, T. In other words, nothing in the formal system prevents this configuration, where V is not a thematic head:

Now, if Voice is introduced with φ-features in the next derivational step, the internal KP would be automatically deactivated (double arrow = Case valuation):

Here, K is invisible for receiving the θ-role from Voice, which would then assign it to a potential external KP, if any. Either way, the VP cannot receive a proper denotation at LF. If the DP denotes in e and K is empty or an identity function, we obtain a type mismatch at LF (cf. also (49)):

This is how we can reinterpret a violation of the Θ-Criterion in the present framework. A KP without θ-role in the relevant domain cannot provide the right denotation for semantic computation at LF. Now, suppose that Voice is φ-defective. Under this circumstance, Voice assigns its θ-role to the internal KP:

(85)	a.	Juan se quejó.
	b.

At LF, this produces a legitimate output. Thus, the relevant interpretation of VoiceP for a sentence like (81a), repeated below, would be as in (86b):

The remaining routine of LF computation until TP is trivial.

Before concluding this subsection, we should wonder whether this approach predicts that all instances of inherent se results in a unaccusative syntax.¹⁶ I think that there is no expectation that this should be the case, since unaccusatives do not have an agentive syntax in the first place. Yet, I do think that this approach would predict a mixed behavior of the subject of inherent se constructions. This seems to be correct. Consider, for instance, the fact that inherent se verbs, like unaccusatives, are incompatible with -dor nominalizations (e.g., trabajador ‘worker’ vs. *quejador ‘complainer’), but, unlike unaccusatives, they cannot participate in absolute constructions (e.g., Llegado Juan… ‘Once Juan arrived…’ vs. *Quejado Juan… ‘Once Juan complained…’).¹⁷

In summary, the proposed system provides thus a principled reason of why inherent se sentences show the transitivity restriction commented above: if Voice valued Case with the internal argument, this argument would remain without a proper denotation and a type mismatch would obtain at LF. At the same time, the theory also explains why clitic insertion is unavoidable here; i.e., why they are inherently pronominal verbs. The reason is that the internal KP receives the θ-role associated with the Voice head, preventing thus the introduction of another KP in Spec,VoiceP. In true reflexive sentences, none of these restrictions show up because V is a thematic head; therefore, Voice can occur with or without φ-features giving rise to the reflexive alternation.

I think that this approach to inherent se generalizes to what Di Tullio (2005) calls “diacritic se”, a “type” of se in which the presence or absence of the clitic changes the basic meaning and valence of the verbal predicate. This set is very broad and, at first glance, there are no systematic meaning connections among the different verbal predicates that combine with this type. For instance, the verb acordar means ‘to agree’ but ‘to remember’ when it combines with the clitic:

The crucial difference with inherent se is that in cases like these there is some sort of competition between the variants with and without the clitic that results in a meaning change. One way of making sense of this competition consistent with the present framework is encoding the difference in the formal makeup of the verbs that participate in this alternation. Thus, I conjecture that the nonpronominal variant enters the syntax with a θ-role that is absent in the pronominal one:

For a transitive verb, the analysis in (88a) has nothing special, i.e., the relation between V and the internal KP is mediated through θ-assignment and its meaning is compositionally determined at LF. In contrast, the relation between V and the internal KP in (88b) is not derived through principles of thematic assignment and semantic composition, but is stored in memory and interpreted by accessing the Encyclopedia. In other words, determination of the meaning in cases like (88b) requires access to arbitrary lists and lot of lexical and world knowledge. One of the most interesting consequences of this way of approaching the issue is that it provides a solution to the problem of metonyminc readings in fake reflexives:

According to Labelle (2008), these constructions constitute a serious challenge for the “unaccusative” hypothesis of reflexives and, at the same time, support her own approach. On her analysis of French reflexives, the clitic se is the surface manifestation of a particular Voice head, which requires an open predicate as argument. Thus, in a sentence like (89), the verb introduces a variable making reference to Juan’s words or actions, which is not syntactically satisfied. As a result, the predicate remains open and can combine with this particular Voice head:

There are two main problems with this analysis. First, the individual variable in (90) comes from a dubious paraphrase of the original sentence, based exclusively on certain meaning intuition connected to the verb meaning. The most obvious meaning of a sentence like (89) is just that Juan is clear. This concrete meaning is exclusively based on some encyclopedic knowledge about the verb at hand and the syntactic environment in which it appears. Second, Labelle’s strategy does not generalize. As she acknowledges, there are other predicates whose meaning cannot be resolved introducing the type of denotation proposed in (90). Consider the following Spanish sentence translated from the French example in (Labelle 2008: footnote 24):

Here, the meaning of tratarse is related to medical care. This is fully unpredictable and, consequently, must be listed in the Encyclopedia. Myriads of verbs participate in the type of alternation we are discussing and, for each case, accessing to lexical and world knowledge is unavoidable.

Now, according to Embick & Marantz (2008), special meanings pertain to the lowest domain of the clause (in their terms, the lowest category-defining head and the Root). If Labelle’s examples are particular instances of inherent/diacritc se, as I am proposing, then we can make sense of the particular meanings that these predicate have when they occur as pronominal verbs. The two sentences in (89) and (91) would then receive the same analysis:

The data discussed so far show not only that Labelle’s argument against the “unaccusative” hypothesis does not hold, but that her theory fails to account for many instances of inherent/diacritic se, including standard cases like quejarse, for which an open predicate analysis does not seem plausible.

5.2 Ergative SE

The sentences in (1), repeated below, have received much attention in the literature (see Schäfer 2008 for extensive discussion and references, and Pujalte 2012 for Spanish in particular):

Unlike impersonals or reflexives, ergatives have no agentive meaning. For this sentence, the sinking event can be related to an internal or an external cause. The con-phrase in (93b) introduces an external cause. Internal cause readings in Spanish can be triggered by adjectives like solo ‘solo’.

Interestingly, ergatives are also compatible with a non-agentive participant of the event. This participant is realized as a dativo de interés, a variety of the ethical dative:

Space reasons prevent me from providing a detailed analysis of se ergative constructions and comparing it with the huge literature on the topic, but some considerations are worth-mentioning. First, the analysis proposed for inherent se in the previous section has obvious consequences for the proper analysis of ergatives. A core ingredient of my approach to inherent se is abandoning Chomsky’s assumption that subcategorization entails θ-marking, an idea already proposed in Postal & Pullum (1988). An inherent se configuration is one in which the verb subcategorizes for a KP that is θ-marked by Voice, not by V. The logic of this system allows for a situation in which V has a subcategorization feature and θ-role but Voice has only a [D]-feature, i.e., it is not a θ-assigner. I would like to suggest now that this logical option is indeed instantiated by anti-causatives:

(96)	a.	El barco se hundió.
	b.

Here, the internal KP receives a θ-role from V and Case from T, since Voice does not have φ-features. The semantic computation at LF is routine. Note only that in this case, Voice does not introduce any θ-role, so it is unique contribution reduces to the identity function it denotes (cf. (54)):

This analysis captures in a direct way the non-agentive property of ergatives. Moreover, it also explains in a simple way the compatibility of se-ergatives with the dativo de interés illustrated in example (95). The reasonable strategy would be adopting an applicative analysis for this special dative:¹⁸

(98)	a.	A Juan se le quemó el asado.
	b.

In principle, it seems that the theory captures in a very similar way the poorly understood alternation among unacusatives predicates with and without se:

In both cases, Voice only denotes an identity function, but in the se variant, Voice also comes with an inherent [D]-feature. Again, this is the perfect counterpart of inherent se. It seems that dissociating subcategorization and θ-marking provides an explanation for the existence of this kind of “deviant” cases.

Truth-conditionally, this alternation does not produce any variation, but there is, however, a subtle difference in meaning for each member of pair in the alternation. For instance, as noted by Pujalte (2012) and others, the sentence in (99b) is incompatible with the dead being provoked by a volitional agent.

The casuistic is extremely complicated (making reference to the cause of the event, to the truth of the propositional complement, to the source of the event, and so on), but it seems that there is a very abstract property in common for each pair, namely, the variant with se introduces a presupposition regarding the semantic nature of the internal argument of the predicate involved in the sentence. This idea can be implemented in the present system by stipulating that the variant with se comes with a presupposition in Voice. Putting the idea in a more technical way, I conjecture that se-Voice denotes a partial identity function, whereas Voice lacking [D] is just an identity function:

Thus, both sentences lack an agent θ-role, but the version with se has particular syntactic and semantic properties: because of the probe nature of se, it triggers A-movement in the syntax but also induces a presupposition in the semantics.

5.3 Aspectual/Benefactive SE

Recall the basic alternation that gives rise to the so-called aspectual/benefactive se:

A high applicative analysis, along the lines of Labelle’s approach, with the corresponding modifications, seems to be a good analytical option, which fits the expectations of my analysis. Here is a tentative analysis:¹⁹

The applied argument receives two θ-roles, one from Appl and another one from Voice. Minimality under A-movement is obviously satisfied, but it remains to be seen why Activity is satisfied as well. This is connected to the general question about Case assignment for some applied arguments. In Spanish, they are syncretic with the dative Case in ditransitives and in other related environments. I assume here that the benefactive values its Case with a probe above Voice. Alternatively, if a competition-based approach is assumed, we can see that at the phase level (the VoiceP) Juan is in the position of getting nominative not dative, which means that it was active in a previous derivational step.

Of course, the tree in (104) is an oversimplification; in particular, it says nothing related to well-known facts about the behavior of direct objects in these aspectual/benefactive se environments (for instance, the incompatibility with bare objects). But I think that these issues are orthogonal to my point here and, for this reason, I leave them open for future investigation. I also leave open discussion with other alternatives that are compatible with my general theory, as well. In principle, my analysis seems to be consistent with Campanini & Schäfer (2011), according to which aspectual se is not derived from a high applicative syntax but of a low one.

The considerations made in this section had the main goal of making explicit how to proceed methodologically given any occurrence of the clitic se in Spanish. The crucial methodological step is this: unless empirical evidence dictates otherwise, assume that se is a probe for A-movement. Given the θ-system proposed here, there is a priori no reason to suppose that an agent KP is also a Voice specifier. Yet, that this is or is not the case is a purely empirical matter. For the se constructions analyzed in this paper, this methodological move has brought the important empirical result of making sense of the apparent chaos in the realm of se constructions. Showing that this strategy can be generalized requires a case-by-case study.

5.4 Summary: Voice deconstructed

Table 1 summarizes the formal aspects regarding both the formal content of Voice and the formal content of the KP that establishes an Agree relation with the clitic and a thematic relation with Voice, in those cases in which there is indeed such a KP in the derivation.

Table 1

Voice deconstructed.

What is se, then, under the present proposal? It is a D-clitic that satisfies the subcategorization requirement of Voice. Since it is defective in the sense that it does not project K, the clitic cannot be a receptor of θ-roles. Table 1 shows that a subcategorization [D]-feature in Voice is what all instances of se have in common. Thus, we derive the presence of se as syntactic expletive. Now, se is also a probe, and as such can attract a KP, which is able to receive the agent θ-role, if there is such a role in Voice. According to the present theory, all the syntactic environments in which there is an active KP result in an A-chain with the moved KP, the clitic and the trace of KP as the members of the chain. Again, this is what all instances of the so-called paradigmatic se have in common, namely: a derived subject, which is born as an active KP within the complement domain of Voice. This is the unaccusative part of this kind of u-syncretism.

Thus, simple reflexive sentences like (4) instantiate a case in which Voice has a [D]-feature and an agent θ-role but no φ-features. In this scenario, the internal KP moves to a position from which it values the φ-features of the clitic, deletes the EPP feature the clitic has, and receives a θ-role from Voice:

It could be the case that Voice has also unvalued φ-features and enters into an Agree relation with some direct object, but there is still another active KP in the VP domain. This is the case of reflexivization of goal KPs in ditransitive constructions (cf. ex. (44a)):

Because of Minimality, an object KP cannot be reflexivized in the same double object contexts (cf. (35b)):

There are cases in which V subcategorizes a KP although it does not θ-mark it. In those cases, such an argument can be properly interpreted if (i) Voice has a unassigned agent θ-role (because of the defective nature of se), but (ii) it does not have φ-features. This scenario results in the so-called inherent se (cf. (7)):

As proposed in many works on anti-causatives (see Schäfer 2008 for an original proposal), Voice could be also semantically empty, but still subcategorize for a nominal. If se is inserted, then the clitic itself satisfies this subcategorization requirement and, in addition, triggers A-movement of the theme KP. Note that it is necessary that V selects such a KP and also θ-marks it, otherwise, the argument cannot be properly interpreted at LF. This is the so-called ergative or anti-causative se (cf. (1)):

The set of paradigmatic clitics discussed to some extent in the present study is completed by the so-called aspectual/benefactive se, which is a case in which Voice combines with se (keeping then its θ-role) but also has φ-features. If there is an added benefactive in its domain, then se attracts the benefactive, which agrees with it, and receives the agent θ-role from Voice (cf. (5)):

Whenever the clitic fails to attract a KP, either because such a KP is simply not there (e.g., impersonal se coming from unergatives verbs, cf. (111) below) or because there is such a KP but with a Case feature already valued (e.g., passive and impersonal se coming from transitive predicates; cf. (2) and (3)), we obtain the so-called non-paradigmatic se:

In these situations, the agent θ-role in Voice remains unassigned in the syntax. At LF, this θ-role is then realized on Voice itself, which takes se as an indefinite argument. As discussed in section 3.1.2, the difference in agreement between impersonal and passive se boils down to a different mechanism of agreement resolution at PF, as proposed in Pujalte (2012; 2020), Pujalte & Saab (2014) and Ormazabal & Romero (2020).

Thus, the present theory conceives of most, if not all, instances of the clitic se and its agreeing variants as triggers for A-movement, i.e., formal probes. There is no need for specific se constructions. There is only one expletive se acting as a syntactic probe. Whenever A-movement succeeds the computational component generates what we superficially call “paradigmatic” se sentences. Whenever Agree fails, because there is no A-movement, the system generates what we superficially call “non-paradigmatic” se sentences.

The theory also dispenses with Voice features of any type (e.g., Active vs Non-Active). The syntax only manipulates three types of formal objects: subcategorization features, θ-roles and φ-features. Such material is manipulated under well-known restrictions, namely, Activity and Minimality.

There are of course many remaining issues opened by the present research agenda. As noted by a reviewer, we should explore what type of combinations of the formal features discussed so far are really present across languages.²⁰ Moreover, if we really want to eliminate Voice features entirely, we should integrate analytical passives into the picture. As suggested in Pujalte & Saab (2012) and Saab (2014), the analytical passive would be a case in which there is an agentive Voice head present in the derivation, but no subcategorization feature in the same head and, thus, no clitic.²¹ Given that, by hypothesis, a thematic head must have a category feature in order to be a θ-role assigner, there is no agent θ-role assignment in analytical passive contexts and Voice is realized as having an existentially closed agent. This would explain why, for instance, there is no pronominal binding of a possessor in cases like the following one, which, as shown by MacDonald (2017a), minimally contrasts with the example (78), repeated below:

This is, in a nutshell, how the present research agenda should be expanded in the future. I think that the agenda is interesting since it opens our analytic space in some intriguing ways. If the theory is at least partially correct, we should push the deconstruction project even further. I cannot explore such additional research routes here, but before closing this article, I would like to provide some final conjectures regarding the question of how some clitics, which are born as regular variables, become formal probes.