P-omission in ellipsis in Spanish: Evidence for syntactic identity

Abstract

In this paper I discuss apparent violations to the P(reposition)-Stranding Generalization (Merchant 2001) in Spanish, a language that does not allow P-stranding in regular wh-questions. I will refer to these apparent violations to the P-stranding Generalization as cases of P(reposition)-omission, which I define as the omission of a preposition in an ellipsis fragment. In order to provide a unified analysis for all types of clausal ellipsis, I examine different constructions such as sluicing, fragment answers, contrast sluicing, stripping and pseudostripping, split questions, and sprouting. I claim that P-omission in clausal ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move. I account for the distribution of this phenomenon by arguing that ellipsis is licensed under strict syntactic identity, and that the remnant doesn’t need to move to escape deletion. Additionally, I show that previous approaches that derive P-omission in Spanish from non-isomorphic, copular sources make incorrect predictions with regard to the patterns found in different types of clausal ellipsis in this language.

1 Introduction

Merchant (2001), based on a survey of more than twenty languages, argues that Preposition-stranding under sluicing is allowed only in languages that allow P-stranding in regular wh-questions. This claim is known as the P(reposition)-Stranding Generalization (1). For reasons that will become clear soon, I’ve replaced “P-stranding” with “P-omission” from Merchant’s original version of (1). In particular, I define P-omission as the omission of a preposition in an ellipsis fragment:¹

• (1) A language L will allow P-omission in sluicing iff L allows preposition stranding under regular wh-movement.

  (adapted from Merchant 2001: 92)

Merchant provides data from four Romance languages (Catalan, French, Spanish, and Italian) and notices that there is a difference in the grammaticality status of P-omission in the sluicing examples, such as Spanish (2a), which are characterized as deviant but not entirely ungrammatical, compared to P-stranding in regular wh-questions (2b), which is undoubtedly ungrammatical in the four Romance languages analyzed:

• (2)

  

  (adapted from Merchant 2001: 98)

In recent years, compliance with the P-Stranding Generalization has been explored for different languages (see e.g. Fortin 2007 and Sato 2011 for Indonesian; Hartman 2005 for Finnish; Szczegielniak 2008 for Polish; Stjepanović 2008 for Serbo-Croatian; Algryani 2010 for Arabic; Abels 2017 for Bulgarian; Molimpakis 2019 for Greek, among others; for Romance languages, see e.g. Almeida and Yoshida 2007 for Brazilian Portuguese; Rodrigues et al. 2009 for Brazilian Portuguese and Spanish; and Vicente 2008 for Spanish). In particular, Rodrigues et al. (2009) analyze the interaction between sluicing and P-omission in Spanish and Brazilian Portuguese;² according to them, P-omission in sluicing in Spanish is only slightly marginal or even totally acceptable, as shown in (3):

• (3)

  

To account for the data in Brazilian Portuguese and Spanish, they claim that the P-less version of the remnant arises from a non-isomorphic copular source: the remnant—cuál ‘which’—is actually the pivot of a cleft/copular sentence that undergoes ellipsis, as schematically shown in (4) (for further details on this proposal see Rodrigues et al. 2009):

• (4)

  

However, an interesting contrast arises in Spanish when comparing the availability of P-omission in sluicing (2a–3) with other types of clausal ellipsis, such as fragment answers, a type of clausal ellipsis in which the answer to a question is stated as a fragment instead of a full sentence (Merchant 2005).³ Merchant notes that ‘bare’ DP answers (i.e. P-less remnants) are impossible in non-preposition stranding languages, extending his P-Stranding Generalization to this type of clausal ellipsis. He provides data from Greek, German, Yiddish, Czech, Russian, Bulgarian, and Hebrew. In Spanish, P-omission is impossible in this type of ellipsis as well, as predicted by Merchant:

• (5)

  

This contrast between sluicing and fragment answers with regard to the availability of P-omission observed in Spanish seems hard to explain within current proposals that argue that counterexamples to the P-Stranding Generalization are derived from non-isomorphic copular sources (see e.g. Vicente 2008; Rodrigues et al. 2009; Barros 2014; among others), given that a copular source for the fragment answer in (5B) is available as a non-elliptical answer, as (6) shows:

• (6)

  

If P-omission in clausal ellipsis arises from copular sources, and if copular sources are just another available source for the ellipsis site, then the ungrammaticality of P-omission in fragment answers in Spanish remains unexplained.

To sum up, on the one hand, both sluicing and fragment answers are types of clausal ellipsis and are derived from the same mechanism of TP-deletion; on the other hand, while sluicing allows P-omission (3) in Spanish, fragment answers do not (5), despite the fact that a copular continuation is perfectly possible (6). Ideally, there should be a way to account for this difference without proposing construction-specific mechanisms that apply in one case but not in the other, given that there are no independent reasons to introduce such differential treatment. In this paper I develop a proposal that accounts for this puzzle in a simple way, also predicting the patterns found in different types of clausal ellipsis.

In a nutshell, against previous proposals that claim that P-omission in Spanish arises from copular sources and that ellipsis is licensed under semantic identity (Vicente 2008; Rodrigues et al. 2009; Barros 2014), I argue that (i) clausal ellipsis is licensed under strict syntactic identity, and (ii) that the remnants can stay in situ—given that, in the relevant cases, the head that would trigger movement in non-elliptical cases does not trigger it under ellipsis. This proposal will account for the following generalization, which I motivate in the remainder of the paper:

• (7) The P-(reposition) Omission Generalization for Spanish:

  P-omission in ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move.

Crucially, although data from apparent violations to Merchant’s P-stranding Generalization have been taken as evidence for the existence of non-isomorphic sources and the need for semantic identity, a careful analysis of several types of clausal ellipsis will demonstrate that these data in fact point to the need to posit strict identity between the antecedent and the source, especially with regard to the position of the remnant and the position of its correlate. In addition, in this paper I show that remnants of ellipsis can stay in situ. Finally, I also show that the P-Omission Generalization holds because movement of the remnant’s correlate in the antecedent creates a configuration where, in order to comply with strict syntactic identity, it is necessary to leave the preposition outside the ellipsis site.

The structure of this paper is as follows: In Section 2, I advance a syntactic identity analysis to account for the original puzzle regarding the (un)availability of P-omission in sluicing and fragment answers. In Section 3, I provide further evidence to show that the analysis proposed here makes the correct predictions regarding the (un)availability of P-omission in other types of clausal ellipsis in Spanish: contrast sluicing in Section 3.1, split questions in Section 3.2, fragments in Section 3.3, stripping and pseudostripping in Section 3.4, sprouting in Section 3.5, and the interaction between sluicing and fragment answers in Section 3.6. In Section 4, I discuss some differences between Spanish and English, and the locus of crosslinguistic variation. Finally, Section 5 discusses the general consequences of the proposal put forth here and some open issues, and concludes.

2 A syntactic identity approach to clausal ellipsis

As I pointed out above, in Spanish, P-omission is possible in sluicing but not in fragment answers. While copular sources could explain the sluicing facts, they cannot explain the impossibility of P-omission in fragment answers. For other arguments against an analysis of P-omission based on copular sources see Stigliano (2019). In that paper, the author presents evidence to rule out copular/cleft sources as the source of P-omission in clausal ellipsis in Spanish. This evidence comes from the availability of non-exhaustive readings (given by mention-some and else modification) in clausal ellipsis with P-less remnants. In this Section I put forth a proposal that accounts for basic patterns found with regard to the (un)availability of P-omission in Spanish in a uniform manner. In Section 3 I show how this proposal correctly predicts the patterns found in other types of clausal ellipsis, such as contrast sluicing, fragments, stripping, pseudostripping, split questions, and sprouting.

2.1 Deriving the P-Omission Generalization

When comparing the two relevant examples (3) and (5), repeated below in (8),^4,5 it becomes evident that one of the features that distinguishes sluicing from fragment answers is the position of the remnant’s correlate (underlined below) in the antecedent: whereas con alguien ‘with someone’ doesn’t move in sluicing (8a), con quién ‘with who’ has been fronted due to wh-movement in fragment answers (8b):

• (8)

  

I claim that this distinction with regards to the movement/non-movement of the remnant’s correlate is what gives rise to the (un)availability of P-omission in each case; this can be stated as follows:

• (9) The P-Omission Generalization for Spanish (first version):

  P-omission in ellipsis is only allowed when the remnant’s correlate does not move.

The rest of this section will develop a proposal that derives the P-Omission Generalization in Spanish. In a nutshell, I claim that clausal ellipsis is licensed under syntactic identity, and that the remnant doesn’t need to move to escape ellipsis (see Section 2.3 for additional predictions). Importantly, I argue that these components—i.e. the strict syntactic identity condition to license ellipsis and the lack of movement of the remnant—are what give rise to the availability of P-omission in some types of clausal ellipsis in Spanish.

In what follows, I describe the main intuition behind this proposal, as a preview of the specific implementation offered in the Section 2.2. Consider sluicing in (10) first, where the underlined text represents the portion of the antecedent taken into account for the identity condition, and strikethrough text represents what’s being elided:

• (10)

  

(10a) illustrates a case in which the entire PP con quién ‘with who’ survives ellipsis, hence, there is no P-omission. What’s being deleted here (i.e. Sonia habló ‘Sonia talked’) is syntactically identical—in a way to be defined soon—to the relevant portion of its antecedent (which is underlined). Likewise, (10b) illustrates a case where the preposition falls inside the ellipsis site, giving rise to P-omission. Here, again, what is being deleted (i.e. Sonia habló con ‘Sonia talked with’) is identical to the relevant portion of its antecedent. What’s crucial in these examples is that the remnant doesn’t move; in consequence, the preposition can optionally fall inside the ellipsis site. Although this proposal seems to imply that (10b) is a case of non-constituent deletion, I will soon show that this is only an illusion.

In fragment answers, on the other hand, movement of the wh-PP Con quién ‘with who’ in the antecedent prevents the preposition from being deleted, as (11) shows:

• (11)

  

(11a) illustrates a case in which the preposition doesn’t fall inside the ellipsis site; this is derived as (10a) above. However, if the preposition falls inside the ellipsis site, as (11b) illustrates, the identity condition won’t be met, given that what’s being elided—habló Sonia con ‘talked Sonia with’—and the relevant portion of the antecedent, which is underlined—wouldn’t be identical (i.e. the latter is lacking the preposition).

To sum up, the P-Omission Generalization holds because movement of the remnant’s correlate in the antecedent creates a configuration in which it is necessary to leave the preposition outside the ellipsis site to comply with a syntactic identity condition. In the remainder of this section I will put forth a way to implement this analysis. Then, in Section 3 I expand the empirical domain to other subtypes of clausal ellipsis, showing how the patterns found regarding P-omission in Spanish fall under the P-Omission Generalization.

2.2 A possible implementation of a strict syntactic identity condition

Moving on to the explicit implementation I propose in this paper, I claim that ellipsis is triggered by [E], and licensed by a special feature, which I represent as [†] (following the notation in Müller 2011). In particular, the [†]-feature is responsible for (a) deletion of any head that bears it, and (b) the licensing of ellipsis. Here I follow previous work (see Saab 2008; Aelbrecht 2011; Murphy 2016; Saab and Lipták 2016; among others) in taking ‘deletion’ to be non-application of Vocabulary Insertion (VI) (Halle and Marantz 1993, and subsequent work) for any head bearing [†], which follows from the principle in (12):

• (12) Phonology of [†]: A head containing the feature [†] is not subject to VI.

The [†]-feature is assigned as in (13) (see Saab 2008, 2010, in press, on I-Assignment for a similar proposal). The constraint on the assignment of [†] to any [F]-marked constituents follows naturally from the assumption that [F]-marked constituents cannot be deleted:

• (13) Assignment of [†]: Assign [†] to every head h in the complement of a head $z_{{[}\mathrm{E}{]}}$ iff h is not dominated by an [F]-marked node.

I follow Merchant (2001) in his assumption that ellipsis is licensed in the complement of heads bearing the feature [E].⁶ Furthermore, as I pointed out above, I claim that ellipsis is licensed under syntactic identity (for additional arguments for syntactic identity, see Chung et al. 2006; Chung 2013; Merchant 2013; among others). As a way of implementing this claim, I adopt a head-by-head evaluation of the identity condition (see Saab 2008, 2010, in press; Tanaka 2011; and Rudin 2019 for similar proposals):

• (14) Identity Condition: A head $h_{[\dagger ]}$ is licensed iff h has an identical correlate $h'$ in A, where A is the antecedent.

In particular, here I adapt Saab’s (2008, 2010) definition of identity:

• (15)

  

Following Saab (in press), I assume that the calculation of ellipsis proceeds top-down, that is, identity is calculated in turn for each head in the E-site, starting from the top-most head that has been [†]-assigned, and that identical heads must be in the same structural position to license ellipsis. Given that the [†]-feature has consequences for Vocabulary Insertion, I claim here that the identity condition should be evaluated in the syntactic component of the grammar, before Spell-Out.

I now move on to briefly illustrate how the current proposal derives a simple case of a fragment answer with a DP remnant (16). For ease of exposition I present a simplified version of trees, leaving aside some details like head movement of V to v to T, movement of subject, etc. The trees on the left are meant to represent the antecedent, and the trees on the right include the ellipsis site and the remnants (i.e the pre-elided clause or source). As the structures below show, I use ( __ ), which signals the base position of the moved constituents, to simplify the trees, although I acknowledge that the same results could be obtained using copies, under the assumption that only the highest copy can serve as a proper correlate to calculate identity. This derives the fact that whatever undergoes movement in the antecedent won’t be able to serve as a correlate for a given head in the E-site, following the original observation by Thoms (2015) that a trace cannot be the antecedent for a non-trace (see also Potsdam 1997 for a similar proposal):

• (16)

  

In (16), every head that bears a [†]-feature in the tree on the right has an identical correlate in the tree on the left (i.e. the Antecedent); recall that heads that are dominated by an [F]-mark cannot be assigned [†]. For the sake of explicitness, and following Saab (in press), in (17) I list the identity reference sets of each head bearing [†]. The order in which the heads appear below is the order in which identity is calculated. The subscripts ‘E’ and ‘A’ are only meant to identify heads belonging to the E-site and to the Antecedent respectively:

• (17) Identity reference sets: {〈T_E, T_A〉, 〈Sonia_E, Sonia_A〉, $\langle v_{E}$, $v_{A}\rangle$, 〈compró_E, compró_A〉}

Crucially, the remnant un libro ‘a book’ is not considered when calculating identity because it’s not assigned [†] due to being [F]-marked. Importantly, the claim that [F]-marked material cannot be [†]-marked dispenses with the need of positing (exceptional) movement of the remnant to escape ellipsis, as proposed for fragment answers by some move-and-delete approaches (see e.g. Merchant 2005; Weir 2014). Given my proposal above, movement is not necessary for the remnant to escape the ellipsis site at all. In cases of sluicing, this is derived by the assumption that the C head can optionally bear an [E]-feature or a [wh]-feature in Spanish, but not both. When the elliptical C only bears a an [E]-feature, ellipsis is triggered, but the remnant stays in situ (I assume that only the features on the probe must be checked).⁷ This in-situ analysis makes important empirical predictions, which are discussed in Section 2.3. A third option is logically possible, that is, when both features are present; this is discussed in Section 4 for English. For the time being, I will only discuss the cases in which the elliptical C bears only an [E]-feature in Spanish. However, this doesn’t mean that the remnant cannot move at all in this language; as I will show in Section 3.4, if movement is triggered by a feature located higher than C (e.g. on Top), the remnant will indeed move.

P-omission in Spanish arises from configurations in which the preposition falls inside the ellipsis site. Crucially, when only the DP/wh-phrase is [F]-marked, the preposition ends up being assigned [†], as shown in (18):

• (18)

  

In this case, the Identity Condition is satisfied just as illustrated above. The identity reference sets are provided in (19):

• (19) Identity reference sets: {〈T_E, T_A〉, 〈Sonia_E, Sonia_A〉, $\langle v_{E}$, $v_{A}\rangle$, 〈habló_E, habló_A〉, 〈con_E, con_A〉}

To reiterate, here the preposition con ‘with’ is assigned [†], hence it must find an identical correlate, which it does. Given that con alguien ‘with someone’ hasn’t moved, the preposition con ‘with’ in the Antecedent serves as its identical correlate.

To derive a case of sluicing in which the preposition is spelled-out along with the DP (i.e. no P-omission), as in (10a), repeated in (20), I claim that both the wh-word and the preposition P are [F]-marked. This prevents the preposition from being [†]-marked:

• (20)

  

To obtain an [F]-marked P I assume, following Büring (2016) (see also Selkirk 1995; Büring 2011; among others), that Focus can project horizontally. The relevant projection rule for the purposes of this paper is stated in (21):

• (21) Horizontal Focus Projection: [F]-marking of an internal argument of a head licenses the F-marking of the head.

  (adapted from Büring 2016: 77)

This is illustrated in (22) for the PP con Bruno ‘with Bruno,’ although I assume that the same happens for PPs that include wh-words like con quien ‘with whom’ above. Büring’s Question-Answer Congruence rule states that “[i]n an answer to a constituent question, the element corresponding to the wh-phrase in the question must be a focus” (Büring 2016: 12). This means that, for instance, given the question Con quién habló Sonia? ‘Who did Sonia talk with?’ [F]-marking in the question will correspond to the wh-phrase, and [F]-marking in the answer will correspond to the DP. However, given the Horizontal Focus Projection rule (21), [F]-marking can either target the DP (22a), which corresponds to the wh-phrase, or the PP dominating it (22b):

• (22)

  

To prevent [F]-marking from continuing to project, for instance, to the entire VP, Büring (2016) proposes the following condition:⁸

• (23) Maximize Background: In any tree, maximize the number of (non-synonymous) constituents that are in the background.

  (Büring 2016: 92)

Where being in the background roughly means being given.⁹ I assume here that [F]-marking the prepositions in these configurations comes for free, but that focus projection further up the tree would incur in a violation of (23).

Going back to the derivation of (20), repeated in (24a), the structure below shows how ellipsis targets the material in the complement of C, except for the preposition con ‘with’ and the wh-word quién ‘who’ which are [F]-marked and cannot be assigned a [†]:

• (24)

  

In (24a), as in (16)–(17) above, every head assigned [†] in the complement of C$_{[\mathrm{E}]}$ has an identical correlate in the Antecedent. For the sake of explicitness, (25) provides the identity reference sets. The PP-remnant con quién ‘with who’ is not considered when calculating identity because it’s not assigned [†] due to being [F]-marked.

• (25) Identity reference sets: {〈T_E, T_A〉, 〈Sonia_E, Sonia_A〉, $\langle v_{E}$, $v_{A}\rangle$, 〈habló_E, habló_A〉}

Moving on to deriving the patterns of P-omission found in fragment answers, recall that this type of ellipsis bans P-omission because the movement of the remnant’s correlate in the antecedent makes it necessary to leave the preposition outside the ellipsis site to comply with the syntactic identity condition proposed above. This derives the fact that whatever undergoes movement in the Antecedent won’t be able to serve as an identical correlate for a given head in the E-site. This idea follows Thoms’s (2015) observation that a trace cannot be the antecedent for a non-trace (see also Potsdam 1997 for a similar proposal).

Crucially, when [F]-marking doesn’t project horizontally, the preposition ends up being assigned [†]. In fragment answers, where the remnant’s correlate has moved, a [†]-marked preposition is not licensed, as it fails to find an identical correlate in the Antecedent. This is because the remnant’s correlate (i.e. the PP con quién ‘with who’) has moved, so the preposition con ‘with’ in the Antecedent won’t be matched with con ‘with’ in the E-site. Hence P-omission is ruled out. This is illustrated in (26):

• (26)

  

As the structure above shows, the preposition con ‘with’ in the E-site doesn’t find an identical correlate, given that the preposition con ‘with’ in the Antecedent is not in the same structural position, correctly predicting the ungrammaticality of P-omission. As (27) shows, the identity reference sets are not complete:

• (27) Identity reference set: {〈T_E, T_A〉, 〈Sonia_E, Sonia_A〉, $\langle v_{E}$, $v_{A}\rangle$, 〈habló_E, habló_A〉, 〈con_E, $\varnothing _{A}\rangle$}

The only grammatical possibility for fragment answers is illustrated below in (28b). In this case, [F]-marking projects horizontally and the preposition ends up being [F]-marked, hence it cannot be assigned a [†]:

• (28)

  

In (29) I provide the identity reference sets, which show that all [†]-bearing heads have an identical correlate in the Antecedent:

• (29) Identity reference sets: {〈T_E, T_A〉, 〈Sonia_E, Sonia_A〉, $\langle v_{E}$, $v_{A}\rangle$, 〈habló_E, habló_A〉}

To sum up, from the comparison between sluicing and fragment answers it’s possible to account for the patterns of P-omission in Spanish stated in the P-Omission Generalization in (9) above: in sluicing the remnant’s correlate stays in situ, which in turn allows the preposition in the Source to fall inside the ellipsis site given that it will find an identical correlate in the Antecedent; in fragment answers, on the contrary, the remnant’s correlate moves, creating a configuration in which the only possible option that complies with the Identity Condition is one in which the preposition is not deleted, giving rise to the unavailability of P-omission in this type of construction.

2.3 In-situ remnants

Before finishing this section, I will present some predictions derived from the claim that elliptical C only bears an [E] feature in Spanish, and not a [wh]-feature (or any other movement-triggering feature). Crucially, this predicts that remnants do not need to move and are interpreted in situ. This prediction is borne out in several domains, such as the licensing of Negative Concord Items (NCIs), reciprocals and binding (some of these tests are based on the argumentation in Weir 2014). Here I provide data from two types of clausal ellipsis: fragment answers and fragments (which are further discussed in Section 3.3). It’s worth noting that these predictions can only be tested in constructions whose non-elliptical counterpart doesn’t involve movement.

In the first place, as the following example shows, when there’s negation, NCIs are only licensed in their base position (30B); that is, they cannot be fronted (30B′):

• (30)

  

However, the claim that remnants stay in situ predicts that NCIs should be able occur as remnants of ellipsis. This prediction is borne out, as the examples from fragments (31B) and fragment answers (32B) show:

• (31)

  

• (32)

  

In both cases, the source is (30B), and not (30B′), showing that certain constituents can be the remnant of ellipsis, although they cannot move in non-elliptical sentences.

Another case that points to the same conclusion comes from reciprocals, in particular cada ‘each’ binding los otros ‘the others’ (33a) in Spanish. Here again, movement gives rise to an ungrammatical structure, as shown in (33b):

• (33)

  

Nevertheless, los otros ‘the others’ can occur as a remnant of ellipsis, both in fragments (34B) and in fragment answers (35B):

• (34)

  

• (35)

  

So far I have provided examples in which the DPs cannot move but can nevertheless occur as remnants of ellipsis. In addition to this, further evidence for an in-situ analysis of remnants comes from certain configurations in which the DPs must move. Crucially, these DPs cannot occur as remnants of ellipsis. An example is provided below for a case of binding (36). As (36a) shows, cada estudiante ‘each student’ cannot be bound in its base position, but it can be bound once it moves, as (36b) shows:

• (36)

  

Crucially, the elliptical version is ruled out, as shown in (37), matching the pattern in (36a), in which no movement has taken place:

• (37)

  

In other words, if ellipsis required movement, we would expect to see the opposite pattern, that is, the fragment answer in (37B) should have been possible.

2.4 Interim summary

In this section I accounted for the initial puzzle regarding the (un)availability of P-omission in sluicing and in fragment answers in Spanish by proposing an analysis based on strict syntactic identity—implemented as a head-by-head condition—and the availability of the remnants to stay in situ—for which I provided independent evidence in Section 2.3.

3 Further evidence and predictions

In this section I provide further evidence for the P-Omission Generalization (7), repeated below in (38), and I show that the analysis developed in the previous section correctly predicts the patterns found regarding the (un)availability of P-omission in other contexts of clausal ellipsis in Spanish.

• (38) The P-(reposition) Omission Generalization for Spanish:

  P-omission in ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move.

Likewise, I show that an analysis based on non-isomorphic, copular sources (e.g. Vicente 2008; Rodrigues et al. 2009) is on the wrong track, since it incorrectly predicts the two way correlation that (a) if P-omission is possible, a copular source will be available, and (b) if a copular source is available, P-omission should be possible. A summary is provided below in the table in (39):

• (39)

  

Finally, I discuss the second part of the P-Omission Generalization—i.e. P-omission in ellipsis in Spanish is allowed when the remnant doesn’t move—which is justified by the stripping/pseudostripping contrast, and how it’s predicted by the analysis I propose here.

3.1 Contrast sluicing

The term contrast sluicing is introduced in Merchant (2001) to refer to a type of sluicing in which the wh-phrase contains contrastive material. This is exemplified in (40) for Spanish:

• (40)

  

According to the P-Omission Generalization (first version) in (9), cases of contrast sluicing in which the remnant’s correlate in the antecedent stays in situ should allow P-omission in Spanish. As (41a) shows, this is indeed the case. Crucially, in these examples, a copular source is not available (41b–41c), providing evidence against a non-isomorphic analysis:

• (41)

  

Interestingly, some cases of contrast sluicing involve movement of the remnant’s correlate, as in (42) (also (40b) above). The proposal developed here predicts that whenever the remnant’s correlate moves, P-omission will be impossible. Again, this prediction is borne out, as shown in (42):

• (42)

  

Despite the fact that a copular continuation for (42) is ungrammatical,¹⁰ the availability of a copular source predicts P-omission to be acceptable in (ia), yet it is not. By contrast, the theory developed here predicts that P-omission will be impossible, due to the movement of the remnant’s correlate. This provides additional evidence against non-isomorphic sources:

• (43)

  

In short, two different configurations for the same type of ellipsis (i.e. contrast sluicing) pattern like sluicing and fragment answers, in that only when the remnant’s correlate in the antecedent stays in situ is P-omission allowed. I showed that a non-isomorphic account based on copular sources would make incorrect predictions, but most importantly, I showed the analysis and the generalization proposed here can account for this pattern without introducing further stipulations.

3.2 Split questions

Split questions are structures that contain a wh-question part followed by a tag that is separated from the preceding material by an intonation break (Arregi 2010):

• (44)

  

Arregi shows, based on data in Spanish, Basque and English, that the two parts of a split question are independent clauses, and that the tag is the remnant of ellipsis in a non-wh-question (for further details on this proposal see Arregi 2010). In addition, he shows that it is not possible to omit the preposition in split questions in Spanish (45):

• (45)

  

Crucially, this is despite the fact that a copular continuation is possible (46):

• (46)

  

Rodrigues et al. (2009) agree with Arregi (2010) in that the ellipsis in the tag is licensed under syntactic parallelism with the antecedent, but they claim that sluicing and split questions are licensed under different identity conditions. However, this solution is undesirable, since there are no independent reasons to posit different licensing mechanisms for two types of clausal ellipsis. Under the account proposed in this paper the ungrammaticality of P-omission in split questions is expected: the remnant’s correlate in the antecedent moves, predicting the unavailability of P-omission, in the same way as fragment answers.

In addition, Rodrigues et al. (2009) observe that P-omission is obligatory whenever the antecedent is a cleft-based question, as the following example shows:¹¹

• (47)

  

These authors argue that this follows from the assumption that “ellipsis in the tag is licensed under parallelism with the antecedent” (Rodrigues et al. 2009: 16). Therefore, this is not a true case of P-omission in that it doesn’t involve the deletion of a preposition: the ellipsis site should be a copular clause (matching its antecedent), as in (48), which is ungrammatical with a preposition:

• (48)

  

3.3 Fragments

The term fragment refers to constructions in which only one argument survives ellipsis. These can arise from dialogue sequences, as in the following example from Spanish:¹²

• (49)

  

Regarding P-omission in this type of ellipsis in Spanish, the examples in (50) show it is indeed possible, as predicted: in both (50a) and (50b) the remnant can optionally appear with or without the preposition:¹³

• (50)

  

The availability of P-omission in these examples is predicted by the analysis proposed in this paper, given that the remnants’ correlates (i.e. sobre un tema interesante ‘about an interesting topic’ and con un chico ‘with a boy’) do not move. The fragment in (50b) contrasts with the unavailability of P-omission in fragment answers I discussed above. Crucially, the difference between these two configurations is related to the position of the remnant’s correlate in the antecedent (i.e. a moved one in fragment answers vs. an in-situ one in fragments), and it is not related to the availability of non-isomorphic copular sources (since both fragment answers and fragments allow them).

Additionally, the analysis put forth here predicts that if the remnant’s correlate in the antecedent moves (for instance, due to focus fronting), P-omission will be banned, as in fragment answers. This prediction is borne out, as (51) shows:

• (51)

  

Here, again, a copular continuation would be perfectly possible, which adds further evidence against a non-isomorphic analysis:

• (52)

  

3.4 Stripping and pseudostripping

All the examples provided so far illustrate the first part of the P-Omission Generalization in (53). In this section, I will motivate the second clause of this generalization with evidence from stripping and pseudostripping.

• (53) The Preposition Omission Generalization for Spanish (final version):

  P-omission in ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move.

I use the term stripping to refer to coordinated constructions in which all elements from the second conjunct are deleted, except for an argument and the negative particle ‘not’ (Sag and Hankamer 1976).¹⁴ This is illustrated in (54). Crucially, in Spanish, there are two possibilities regarding the order between the remnant and the negative particle (Depiante 2000), unlike English, which only allows for one of them (i.e. negation preceding the remnant). Depiante dubbed cases like (54b), in which the remnant precedes negation, pseudostripping:

• (54)

  

These two configurations give rise to different patterns regarding the availability of P-omission, in that only stripping allows it:¹⁵

• (55)

  

Interestingly, the non-isomorphic approach would predict the exact opposite pattern, given that a copular source is available for the pseudostripping example in (55b), as (56b) shows, but not for stripping example in (55a), as (56a) shows:¹⁶

• (56)

  

Although the unavailability of P-omission in (55b) constitutes a counterexample to the earlier version of the Preposition Omission Generalization as stated in (9) (i.e. P-omission in ellipsis in Spanish is only allowed when the remnant’s correlate does not move), the analysis proposed in this paper can actually account for these cases. I claim that this difference regarding the availability of P-omission in stripping (55a) and pseudostripping (55b) is due the to the position of the remnant with regard to the negative particle. In particular, Depiante (2000: 127) argues that pseudostripping is derived via movement of the remnant “to a higher functional head F, encoding contrastive focus,” as in (54b), repeated in (57b). This movement is required to obtain the desired word order in pseudostripping (57b), but not in stripping (57a) (recall that according to the analysis put forth in Section 2, the remnant doesn’t need to move):

• (57)

  

To obtain P-omission in pseudostripping it would be necessary to posit movement of the DP remnant over the negative particle, leaving the preposition stranded inside the ellipsis site, which is independently banned in Spanish. I claim that this movement is what causes the ungrammaticality of P-omission in pseudostripping, motivating the second part of the P-Omission Generalization in (53) above:

• (58)

  

Before finishing this subsection, I’d like to add two additional points. First, as I discussed for fragments in Section 3.3, my analysis predicts that in cases of stripping where the remnant’s correlate moves, P-omission will be banned. This prediction is borne out, as shown in (59), where the PP Con Bruno ‘with Bruno’ is fronted:

• (59)

  

The example in (60) also points to the same direction:¹⁷

• (60)

  

Finally, Merchant (2001) observed that when the remnant is fronted in sluicing in Spanish, the preposition cannot be omitted (61b). This contrasts with the non-fronted cases of sluicing like (8a), repeated below in (61a), which optionally allow P-omission:

• (61)

  

If these cases are derived as pseudostripping (i.e. via fronting of the remnant), the unavailability of P-omission is accounted for by the unavailability of P-stranding in this language. For the sake of completeness, a copular continuation for (61b) would be perfectly possible, providing further evidence against a copular source analysis of P-omission:

• (62)

  

3.5 Sprouting

Sprouting is a type of sluicing in which the remnant’s correlate is an implicit argument or adjunct (Chung et al. 1995). Chung (2006) noted that P-omission is impossible in sprouting even in P-stranding languages like English (original observation by Rosen 1976). In Spanish, P-less remnants are also ruled out:

• (63)

  

To account for the impossibility of omitting the preposition in sprouting, Chung proposed a condition, which stated that “every lexical item in the numeration of the sluice that ends up (only) in the elided IP must be identical to an item in the numeration of the antecedent CP” (2006: 11); this condition was further implemented in other proposals such as the one by Rudin (2019), cited before. The formalization proposed here, partially based on Saab (2008), states that the preposition can only be deleted if it’s [†]; if this happens, it must find an identical correlate in the antecedent. Crucially, there is no such identical correlate (in line with the original proposal by Chung 2006), hence the deletion of the preposition gives rise to ungrammaticality.

3.6 Interaction between sluicing and fragment answers

An interesting prediction made by the analysis put forth here is related to the interaction between sluicing and fragment answers (64). What the example below shows is that when a sluiced question allows for the preposition to be deleted, the fragment answer to it also allows P-omission (contrary to what happens in fragment answers to full wh-questions, as discussed before):

• (64)

  

In (64A1) the remnant’s correlate con alguien ‘with someone’ in the antecedent is in situ, which makes P-omission in B’s sluiced question possible, as shown in (64B). This, in turn, would mean that the remnant Quién ‘Who’ in (64B) has remained in situ, as argued above. Furthermore, this predicts that P-omission is possible in a fragment answer to that question. This prediction is borne out, as (64A2) shows. Crucially, this example shows that the (un)availability of P-omission is not dependent on one type of clausal ellipsis or another, but to their syntactic structures and, crucially, the position of the remnant’s correlate.

3.7 Interim summary

In this Section I provided further evidence for the analysis put forth in Section 2, and I expanded the P-Omission Generalization for Spanish. The evidence presented came from different types of clausal ellipsis and the contrasts found with respect to the (im)possibility of omitting the preposition in those contexts. Additionally, I showed how my proposal doesn’t make predictions based on particular ‘types’ of ellipsis, but on the structure that they occur in, as I discussed for the interaction between sluicing and fragment answers. I provided independent evidence for an in-situ analysis of remnants, and I showed I show that an analysis based on non-isomorphic, copular sources incorrectly predicts the two way correlation between P-omission and copular sources, which is summarized in (39), repeated in (65):

• (65)

  

In the next section I will briefly discuss some crosslinguistic differences and how my proposal can account for them.

4 Crosslinguistic differences

As I discussed in Section 2.2, my working hypothesis is that, in principle, the C head can bear an [E]-feature, or a [wh]-feature (or any other movement-inducing feature, which I will call [M]) in Spanish. In Section 2.3 I presented evidence that shows that remnants in Spanish behave as if they do not move. That is, I’ve argued that in Spanish, elliptical C bears only an [E] feature, which makes the remnant stay in situ (unless there’s a movement-triggering feature higher in the structure, as in the cases of pseudostripping in Section 3.4). In this section, I briefly discuss clausal ellipsis in English. I claim that in this language the C head can bear an [E]-feature, an [M]-feature, or both. Crucially, the featural make-up of the C head will provide an explanation for the differences found in English and Spanish with respect to the (un)availability of P-omission in (some contexts of) clausal ellipsis. In addition, in Section 4.2 I briefly discuss how this could account for crosslinguistic differences more generally.

4.1 Spanish vs. English

In English P-omission is optional in fragment answers (66a), which seems to be exactly the opposite of what the proposal in this paper predicts, based on the Spanish data in (66b) discussed so far:¹⁸

• (66)

  

I claim that this difference is due to the featural makeup of the C head. In particular, cases such as the one in (66a) arise from a structure in which the C head that bears both and [E] feature and a movement-triggering feature—which I will refer to as [M] for convenience. Following Müller (2011, 2017), and references therein, operation-inducing features on heads are ordered. Therefore, when [E] and [M] are both present on a given C head, they will be triggered in a specific order. I claim that this order is subject to parametric variation, that is, it can vary across languages (see also Section 4.2). In particular, I claim that [E] is ordered before [M] in English, and this is what derives the difference with respect to the optionality of P-omission in fragment answers between these two languages, illustrated in (66) above.

Specifically, in English the features are ordered [E] ≻ [M], which means that [†]-assignment occurs before movement to the Specifier of the CP is triggered. This gives rise, in principle, to three logical possibilities, summarized in (67). Two of these, i.e. (67a) and (67c), give rise to grammatical structures, and one (67b) is ruled out by the Identity Condition:

• (67)

  

The representation in (67a), in which the entire PP is F-marked and then moves to the left periphery, will generate a PP remnant (i.e. no P-omission); this is further illustrated in (68). First, the [E]-feature triggers [†]-assignment to every head that’s not F-marked in its complement. As the second step indicates, an [M]-feature triggers movement of the PP to the Specifier of the CP. This derivation complies with the Identity Condition:

• (68)

  

A second logical option would be the one in (67b), further illustrated in (69). In this case, the preposition, which is not F-marked, ends up being marked with [†]. Crucially, this derivation would be ruled out, given that it doesn’t comply with the Identity Condition. This is because the preposition in its in-situ position cannot be deleted since it doesn’t have an identical correlate (as already discussed for Spanish throughout this paper):

• (69)

  

Finally, the structure shown in (67c), further illustrated in (70), generates a P-less remnant; this is allowed in this language given that the [E]-feature is ordered first, hence the preposition can be [†]-marked, and then the entire PP moves, pied-piping the preposition. Crucially, here the preposition in the moved PP will find an identical correlate (i.e. the preposition in the moved PP in the Antecedent):

• (70)

  

Crucially, the [†]-marked preposition will find an identical correlate (i.e. the moved preposition in the Antecedent), complying with the Identity Condition.

The hypothesis that elliptical C heads in English can bear either only an [E]-feature, or both [E] and [M] features, could have further consequences such as accounting for the seemingly contradictory findings with respect to the interpretation of fragments. In particular, while Weir (2014) argued that fragments seem to be in situ based on interpretative facts such as NPI licensing, quantifiers, binding, etc., Shen (2018), based on superlative examples, argued that fragments must have moved. Although this is worth exploring in depth, it’s beyond of the scope of this paper.

Unlike English, I argued that C heads in Spanish bear either an [E]-feature or an [M]-feature. It could also be the case that the C head bears both features in Spanish (although this should be supported with independent empirical evidence). In these cases, I claim that the order of these features is [M] ≻ [E] (that is, the opposite of English). Therefore, we would have, in principle, two logical possibilities (71), but only one, i.e. (71a), is possible. On the contrary, (71b) is ruled out given the ban on P-stranding in this language. This option, which doesn’t violate any language constraints and complies with the Identity Condition, gives rise to a remnant with a preposition (i.e. no P-omission):

• (71)

  

The structure in (71a) involves first, the entire PP moving, and then [†]-assignment, triggered by [E]. (72) shows a step by step derivation:

• (72)

  

The second option (71b) would be ruled out in Step 1, given that it involves stranding the preposition, which is banned in Spanish:

• (73)

  

Finally, a structure parallel to English’s (70b), in which the preposition is assigned [†] and then the entire PP moves, will never be generated in Spanish, given that movement is triggered before ellipsis (and, in consequence, the preposition will already be out of the scope of [E]).

4.2 A note on the locus of crosslinguistic variation

One advantage of the proposal I advanced in this paper is that it provides a plausible explanation for crosslinguistic variation. In the first place, I proposed here that there could be variation in the featural makeup of C heads. On the one hand, these differences could be found in which features can be present or absent, and which features can co-occur in a given head. On the other hand, there might be differences in the order in which features are triggered when they co-occur in the same head (e.g. [E] ≻ [M] vs. [M] ≻ [E]). This is the hypothesis I introduced in the previous section to explain the differences between Spanish and English. It’s worth mentioning that this is not just a theoretical conjecture, but it makes empirical predictions as well. As I mentioned, if [E] and [M] are in complementary distribution, then we expect to find data that shows that remnants of ellipsis do not move, as the data presented in Section 2.3 for Spanish. If, on the contrary, we allow for both possibilities (i.e. only an [E] feature or both [E] and [M], we can expect to find seemingly contradictory data, as mentioned with regard to the findings by Weir (2014) and Shen (2018) for English.

In the second place, there could be variation based on the (un)availability of P-stranding in regular wh-questions in a given language. Recall that some P-omission was ungrammatical in Spanish in cases of pseudostripping because of the ban of P-stranding in this language. This means that it should be possible, in principle, to find cases of P-omission in the same contexts in a language that does allow P-stranding (provided that the Identity Condition is met). Some other aspect of the proposal that can explain crosslinguistic variation could be the assignment of Focus, and how the focus projection rules work in each language. However, it’s important to mention that this comment on the locus of crosslinguistic variation does not pretend to be a full theory, but a hypothesis that’s worth exploring, and that should be ultimately supported with empirical data.

Finally, I briefly consider the consequences that the proposal I advanced here have for Merchant’s original P-stranding generalization, repeated in (74):¹⁹

• (74) A language L will allow P-stranding in sluicing iff L allows preposition stranding under regular wh-movement.

  (Merchant 2001: 92)

There are two ways in which this generalization can be interpreted. On the one hand, it can be interpreted so that if a language doesn’t allow P-stranding in regular wh-questions, P-stranding under ellipsis won’t be allowed either, that is, ellipsis doesn’t create a configuration in which the violation that arises from stranding a preposition can be ‘repaired.’ Under this interpretation, I believe that Merchant’s generalization should be maintained, since there’s nothing special in the ellipsis mechanism that allows for exceptional P-stranding in languages that don’t usually allow it. A different interpretation—the one I’ve adopted in this paper, given my modification of (74) in (1)—interprets ‘P-stranding in sluicing’ as ‘P-omission’ (i.e. the omission of a preposition in an ellipsis fragment). In this case, then, Merchant’s generalization is not valid, in that a language doesn’t need to allow P-stranding in regular wh-question to allow P-omission in ellipsis.

5 Further consequences, open issues and conclusions

In this paper I provided data from different types of clausal ellipsis in Spanish that show that some types (namely, sluicing, contrast sluicing, fragments, and stripping) allow P-omission, but some other types (namely, fragment answers, split questions, and pseudostripping) don’t allow it. All these data fall under the generalization in (53), repeated below in (75):

• (75) The Preposition Omission Generalization for Spanish:

  P-omission in ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move.

I also argued against a non-isomorphic, copular source analysis of P-less remnants by showing that there is no one-to-one correspondence between the contexts in which P-omission is available and the contexts in which copular sentences are possible as sources for the ellipsis site. More importantly, I showed that all these patterns follow straightforwardly from the analysis proposed here: I accounted for the distribution of P-omission in Spanish by arguing that (a) whether remnants move or not is construction-specific as implemented by the proposal with varying features on the ellipsis-licensing head, and (b) ellipsis is licensed under strict syntactic identity to/with an antecedent. This provided a unified analysis of all types of clausal ellipsis in Spanish without the need of construction-specific assumptions or stipulations. A very important aspect of the analysis in this paper is that, while most analyses of ellipsis concentrate on the comparison of the Antecedent and the E-site as a whole, I showed that it’s crucial to focus on the specific position of the remnant’s correlate in the antecedent as well (see also Potsdam 1997; Thoms 2015). Finally, I examined some crosslinguistic differences found when comparing languages such as English and Spanish, and I briefly discussed the locus of crosslinguistic variation (which, I claim, is located in the featural make-up of the ellipsis-triggering head C).

Although this paper focused on a very specific empirical zone (i.e. apparent violations to the P-stranding generalization in clausal ellipsis) in a particular language (i.e. Spanish), I believe that the proposal put forth here could have further consequences for the analysis other elliptical phenomena, both in Spanish and in other languages, that are worth examining. In the reminder of this section, I will briefly discuss some of these possible consequences and extensions, although I leave an in-depth exploration of them for future work.

One consequence of my proposal is that it could account for island repair/amelioration effects found in sluicing (first noted by Ross 1969), and other types of clausal ellipsis (see e.g. Culicover and Jackendoff 2005; Fukaya 2007; Valmala 2007 for fragment answers and Potter 2017 for stripping, among many others), in a fairly straightforward way: if remnants do not need to move, no islands would, in principle, arise (for previous proposals that attempted to provide an explanation for this phenomenon see Ross 1969; Lakoff 1970; Chung et al. 1995; Merchant 2001; Culicover and Jackendoff 2005; Müller 2011, among many others; see also Abels 2018 for an overview of this topic). Therefore, it may be possible to eliminate the need of proposing non-isomorphic copular sources, short sources, or exceptional mechanisms to ‘repair’ ungrammatical structures.²⁰

Although this paper discussed the need of an identity condition based on syntactic identity, which is calculated in the syntax, before the derivation is spelled-out, I don’t reject the hypothesis that other types of identity/parallelism are required, on top of syntactic identity, to license ellipsis. In particular, I am agnostic as to whether different identity conditions play a role in licensing ellipsis at different stages of the derivation, and additionally, whether there are other principles that could determine or influence possible interpretations for the ellipsis site (as proposed by Merchant 2010 with respect to pragmatic plausibility, and similarly discussed in Rudin 2019). Since this topic seems more complex and deserves additional research, I will not discuss it any further here, and I leave it open for future work.

Another consequence of the analysis proposed here is related to the (un)availability of non-isomorphic sources (such as copular/cleft sentences) as sources for the ellipsis site crosslinguistically. Even though I argued here that these sources are not the locus of P-omission, and more generally, that they are not available in Spanish, it’s worth exploring other languages in which this type of non-isomorphic source has been proposed to explain certain mismatches such as case or agreement mismatches. For instance, non-isomorphic, copular/cleft sources have been proposed (Barros 2016) to explain the different agreement patterns found in adjectival sluices in languages such as Hungarian, German and Dutch. However, Ronai and Stigliano (2020) claim that a detailed analysis of the phenomenon in Hungarian actually rules out copular sources in that language, and argue for the need of strict isomorphism to license ellipsis. I believe it is necessary to carefully revisit the data taken as evidence for non-isomorphic sources to asses whether it actually points into that direction.

In relation to this last point, the strict identity condition to license ellipsis I proposed here has further consequences for the analysis of (the unavailability of) case-mismatches. In particular, Spanish marks some direct object DPs with Differential Object Marking (DOM) (see López 2012 for a thorough discussion of DOM in Spanish). The prediction here is that, given than DOM is a case-marker, there cannot be DOM mismatches, in line with what happens with respect to case in other languages (although some exceptions have been found—see Vicente 2015 for a brief summary), and following the original observation by Ross (1969). The case-matching effects follow trivially if the sluicing site contains an elided version of the relevant case assigner (here, the verb), and the remnant occupies a position in which it can be assigned the correct case. However, under a hypothesis that deletion of copular sources is just another case of sluicing (like the one I argued against in this paper) it is hard to account for data like (76) below—in which the remnant of sluicing must appear with the DOM a—given that the pivot of clefts bears nominative. In other words, the only possible source for (76) is the wh-question in (77a):

• (76)

  

• (77)

  

Furthermore, Merchant (2001) showed that Ross’s 1969 observation, which he dubbed the Case Matching Generalization, holds even when there is no explicit correlate (that is, in cases of sprouting). In sprouting there is no correlate but the remnant still appears with the correct case marking (i.e. in Spanish, with dom if the object is human/animate, and without dom otherwise). This marking is obligatory (78a), even when a copular source with an unmarked DP is available (78b):

• (78)

  

That is, the data from DOM matching, both in sluicing and sprouting, points to the need of having an isomorphic E-site, and against something like a copular/cleft source.

As for some open issues that deserve further exploration, I should mention multiple sluicing (usually defined as a type of sluicing in which multiple wh-phrases survive deletion). Rodrigues et al. (2009) note that P-less remnants are impossible in multiple sluicing in Spanish (79).²¹ Crucially, the proposal in this paper would predict, in principle, that the prepositions from both PPs could be optionally omitted, contrary to fact:

• (79)

  

Although (79) is indeed ungrammatical with both remnants omitting the preposition, it has been noted by Martín González (2010) that similar examples optionally allow omitting the preposition only for the first remnant, as the following example shows:

• (80)

  

Comparing examples such as (79) and (80) complicates the empirical picture, because it doesn’t seem to be the case that P-omission is always banned in these contexts. Likewise, I must note that multiple sluicing seems to be special in many aspects. For instance, it seems to be the case that P-omission is also unavailable under multiple sluicing in languages with more generalized availability of P-omission, such as English (see e.g. Lasnik 2014). At the moment, I don’t have a clear explanation for why this is the case, so I leave an in depth investigation of this type of ellipsis for future work.

Finally, as an anonymous reviewer points out, although the cleft-source analysis proposed by Rodrigues et al. (2009) faces some empirical challenges for the data in Spanish, as correctly pointed out here and in Stigliano (2019), it does seem to account for P-omission under sluicing in Brazilian Portuguese, where this phenomenon seems to be much more pervasive than in Spanish.

To conclude, in this paper I discussed apparent violations to the P-Stranding Generalization—which I referred to as cases of P(reposition)-omission—in Spanish, a language that does not allow P-stranding in regular wh-questions. I examined different types of clausal ellipsis such as sluicing, fragment answers, contrast sluicing, stripping and pseudostripping, split questions, and sprouting, and I provided a uniform treatment for all of them. I claimed that P-omission in clausal ellipsis in Spanish is only allowed when the following two conditions are met: (a) the remnant’s correlate in the antecedent does not move, and (b) the remnant does not move. I accounted for the distribution of this phenomenon by arguing that ellipsis is licensed under strict syntactic identity. Additionally, I showed that previous approaches that derived P-omission in Spanish from non-isomorphic, copular sources make incorrect predictions with regard to the patterns found in different types of clausal ellipsis. Finally, I advanced a proposal to account for some crosslinguistic differences.