1 Introduction

Gapping is an ellipsis typically found in coordinate structures. While gapping has been extensively studied in the literature, some aspects of its syntax are still poorly understood, in particular with regard to the No Embedding Constraint according to which gapping cannot occur in embedded clauses (Hankamer 1979; Neijt 1979; Johnson 2009; 2017; 2018):

This fact has led researchers to conclude that gapping is a root phenomenon, in contrast to other types of ellipsis such as VP-ellipsis, pseudogapping, fragments and sluicing, all of which can occur in embedded contexts. However, the validity of the No Embedding Constraint is challenged on empirical grounds, since embedded gapping is acceptable in languages like Farsi (2) and Spanish (3) (see Erschler 2018 for a survey of typologically different languages that allow embedded gapping):

This poses a series of questions as to whether the No Embedding Constraint is consistent and robust enough to cope with data like (2) and (3), and if so, how it can be circumvented in languages where embedded gapping is allowed. Previous studies have shown that there is a correlation between the acceptability of embedded gapping and the type of matrix predicates (see Saab 2008; Fernández-Sánchez 2017; Bîlbîie & de la Fuente 2019; Bîlbîie & Faghiri 2022). However, the reason why English differs in this respect from Spanish (or Farsi) is still unknown and needs to be addressed in a way that reflects crosslinguistic variation.

This paper seeks to examine why embedded gapping is allowed in some languages but not in others, and how this variation can be accounted for in a principled way. I show that the possibility of embedded gapping is related to the phasehood of the embedded CP, assuming that the ellipsis site corresponds to a phasal complement. Crosslinguistic variation in phasehood inheritance is shown to determine whether embedded gapping is grammatical across languages.

The paper is organized as follows. Section 2 discusses the No Embedding Constraint and its implications for the analysis of (embedded) gapping. Section 3 examines the possibility of embedding stripping within and across languages, showing that Wurmbrand’s (2017) phase-based approach to embedded stripping can be successfully extended to embedded gapping. Section 4 provides an account of the difference between English and Spanish with respect to embedded gapping, and claims that phasehood inheritance along the lines of Citko (2020) accounts for the possibility of embedded gapping in these languages. Section 5 explores the implications of this proposal for gapping and (pseudo)stripping in embedded clauses in which English and Spanish behave differently. Section 6 concludes the paper.

2 The validity of the No Embedding Constraint

The No Embedding Constraint in (4) is aimed to capture the fact that neither the gap nor its antecedent may be embedded:

Johnson (2017) claims that if an ellipsis is affected by this constraint, it is gapping as defined in (5), with the exception of gapping in Farsi which, as Farudi (2013) shows, does not follow the constraint in (4):

According to Johnson (2017), the No Embedding Constraint may vary from language to language; gapping in Farsi may be subject to a different type of constraint on embedding which seems to be related to movement out of CP (see Farudi 2013 for discussion).

The literature on gapping has also claimed that the No Embedding Constraint is the result of the syntax of low coordination (Toosarvandani 2013; 2016). If gapping involves low coordination below TP, it naturally follows that there is no embedded gap (or antecedent) within their conjunct (Johnson 1996/2004; 2009); coordination at the vP-level does not, and cannot, make room for embedding in such cases. However, the data given in (2) and (3) do not support such a view because of the possibility of embedding within the second conjunct. In fact, a growing body of work suggests that gapping may involve either clausal coordination or low coordination (see, e.g., Repp 2009 for German; Saab 2008 and Centeno 2011 for Spanish; Dagnac 2016 for French; and Potter, Frazier & Yoshida 2017 for English). Evidence of this comes from scope facts (Siegel 1984; Oehrle 1987; McCawley 1993):

Potter, Frazier & Yoshida (2017) claim that the scopal ambiguities in (6) can be captured directly by suggesting that each scope interpretation has a different source; the wide scope reading (6b) originates from small conjunct gapping, as in (7a), whereas the distributive scope reading (6c) comes from large conjunct gapping, as in (7b):

There are also cases where gapping must be analyzed as deletion of constituents larger than just VP. Consider the following example in Spanish, where gapping does not tolerate tense mismatches between the ellipsis site and its antecedent (Brucart 1987; Murguia 2004; Saab 2008; 2010; 2016; Brucart & MacDonald 2012):

This indicates that the category affected by ellipsis is at least TP. Similarly, when gapping occurs in Clitic Left Dislocation (CLLD) contexts, as in (9), the conjuncts must be large enough to host dislocated elements at the left-periphery:¹

The object la revista of the second conjunct is CLLD-ed, given that the ellipsis clause receives the same semantic and information-structural interpretation as its antecedent (Winkler 2013: 470), and that the linear order of the remnants must be the same as that of the correlates (Johnson 2018: 596). In (9), the ellipsis site is arguably TP insofar as CLLD-ed elements in Spanish are adjoined to TP, (10a) (Zubizarreta 1998), or located in dedicated positions in the CP-area, (10b) (López 2009):

The same is true for gapping in clitic doubling contexts, since clitics in Spanish adjoin to T (or a clitic projection above TP) to which the finite verb moves:

Given these facts, it is fair to say that there are indeed two source structures for gapping. If this is true, the very existence of embedded gapping in languages like Farsi and Spanish does not cast doubt on the relevance of the No Embedding Constraint itself. Rather, there must be syntactic variation on the structure of the embedded clause in which gapping occurs; some languages may circumvent the No Embedding Constraint, whereas others may not.

3 Embedded stripping and gapping

Stripping is a type of ellipsis in which everything from a clause is elided except one constituent (Hankamer & Sag 1976; Lobeck 1995; Depiante 2000; Johnson 2018). What is of interest to us is the fact that stripping, like gapping, cannot occur in embedded clauses:

Thus, it is claimed that both elliptical constructions are subject to the same constraint, namely the No Embedding Constraint by which gapping was analyzed either through VP-ellipsis (Coppock 2001) or Across-The-Board (ATB) movement of VP under low coordination (Johnson 2009). However, stripping involves clausal ellipsis (Depiante 2000; 2004b; Merchant 2003; Vicente 2006; Kolokonte 2008; Saab 2008; Ortega-Santos, Yoshida & Nakao 2014), similar to sluicing, which is derived via wh-movement to [Spec, CP] followed by TP-ellipsis (Merchant 2001). Under Merchant’s (2001) [E]-feature approach to ellipsis, example (12a) can be analyzed as follows:²

This kind of analysis, though widely adopted in the literature, does not have anything to say about the possibility of embedded stripping in English. Wurmbrand (2017) reports data like (14) illustrating that embedded stripping is possible when the complementizer is absent:

She claims that the presence or absence of the complementizer plays a key role in the acceptability of embedded stripping, and suggests the following generalization:

To account for the contrasts in (14), Wurmbrand (2017) assumes a dynamic approach to phasehood—the phasal status of a phrase may depend on its syntactic context (Felser 2004; Bobaljik & Wurmbrand 2005; Den Dikken 2007; Gallego & Uriagereka 2007; Bošković 2013; 2014)—and proposes that the possibility of embedded stripping is associated with the phasehood of the embedded CP. Specifically, Wurmbrand (2017) claims that the difference between matrix and embedded stripping lies in how ellipsis is constrained by phasehood in each case, assuming that ellipsis targets the complement of a phase head (Van Craenenbroeck 2004; 2010; Gengel 2009; Rouveret 2012). The derivation of matrix stripping (16a) is illustrated in (16b), where FocP is a phase, and the spell-out domain is TP which corresponds to the ellipsis site:³

Embedded stripping (17a), by contrast, has the structure in (17b), where CP is a phase, and the spell-out domain is FocP which does not correspond to the ellipsis site (i.e., TP):

As for the contrast in (14), Wurmbrand (2017) claims that when the matrix verb takes a CP-complement, the spell-out domain is not TP but FocP, as in (18a). In contrast, when the matrix verb takes a CP-less clause as its complement, FocP is a phase, and the spell-out domain (i.e., TP) corresponds to the ellipsis site, as in (18b):

Thus, under this analysis, whether or not FocP constitutes a phase is crucial for the possibility of embedded stripping in English. As it stands, though, the generalization in (15) does not hold for languages like Spanish; the presence of the complementizer que is required in embedded clauses, regardless of whether ellipsis occurs or not:⁴

Wurmbrand (2017) claims that if a language employs wh-movement to [Spec, FocP] (e.g., Hungarian), embedded stripping is possible even if there is an overt complementizer:

In contrast, if a language employs wh-movement to [Spec, CP] (e.g., English), embedded stripping is not possible when the complementizer that is present (see (18a)). Given the phase-based analysis of embedded stripping in (18b), in Hungarian-type languages Foc is the phase head that licenses the ellipsis of its complement (i.e., TP) despite the presence of an overt complementizer. This would mean that the crosslinguistic variation with respect to embedded stripping lies in the phasal status of FocP below CP, an issue I return to in Section 4.

Last but not least, Wurmbrand (2017) tentatively suggests that the same reasoning could apply to embedded gapping in English, which is acceptable for some speakers when the complementizer is absent:

Although the acceptability judgments for embedded gapping are less clear than for embedded stripping, there is little doubt that the absence of the complementizer does affect the acceptability of embedded gapping (Wurmbrand 2017: 361). Given the similarity in this respect between gapping and stripping, let us consider whether the following generalization holds for embedded gapping:

This means that the No Embedding Constraint will not apply to all cases; in English, embedded gapping is possible only if the matrix verb takes a CP-less clause as its complement (see (18b) above). However, the generalization in (23) cannot apply to Spanish for the same reason as (19); in embedded gapping, the matrix predicate must take a CP-complement headed by the complementizer que:

This holds true irrespective of whether the matrix predicate is factive or non-factive (see footnote 4). Before moving on to the analysis of embedded gapping, I now discuss some complications regarding the mood in the embedded clause and its relation to gapping in Spanish.

It is claimed that factive predicates do not allow embedded gapping, (25a), in contrast to non-factive ones, (25b) (Bosque 1984; Brucart 1987; Saab 2008; Fernández-Sánchez 2017; Bîlbîie & de la Fuente 2019):

However, as noted by Saab (2008), it is not entirely clear whether the ungrammaticality of (25a) is due to embedding per se, or mood mismatches in TP-ellipsis in general; tense, aspect and mood (TAM) in the antecedent clause and the ellipsis clause must be identical. Accordingly, (25a) is ungrammatical because of the TAM mismatch:

This in turn means that the grammaticality of (24b) is attributed to tam matching:

Yet, the issue is not as simple as it seems. Embedded gapping is still impossible under factive predicates even though the verbs in the antecedent clause and the ellipsis clause are strictly identical:⁵

The question that arises here is: what rules out this kind of examples? While this issue needs some attention, it seems plausible that the subjunctive verb cannot serve as an antecedent if it is included in an embedded clause. More generally, the antecedent of gapping cannot be embedded (Hankamer 1979: 19):

In view of this, (28) is ruled out presumably for the same reason as (29). However, it should be noted here that mood mismatches are tolerable in certain elliptical contexts such as pseudostripping.⁶ Brucart (1987; 1999), Depiante (2004b), and Saab (2008) observe that in Spanish, embedded pseudostripping is possible under epistemic predicates, (30), while it is not possible under factive predicates, (31):

Saab (2008) claims that this contrast has to do with whether the projection responsible for the subjunctive mood is included in the ellipsis site or not, which is crucial for structural isomorphism between the elided TP and its antecedent. He proposes that the subjunctive feature of epistemic predicates is encoded in C, so TP does not contain it, as in (32a), whereas the subjunctive feature of factive predicates is in T and thus included in TP, as in (32b):

In this view, the mood mismatch in (30) is illusory, and there is no syntactic mismatch between the ellipsis site and its antecedent. Instead, the contrast between (30) and (31) is due to the interaction between syntax and morphology concerning the subjunctive mood (see Saab 2008 for more details). Though I leave a detailed discussion of this issue for future work, I tentatively suggest that the difference between factive and non-factive predicates in embedded gapping can be explained as follows; in (25a), the subjunctive feature is included in the ellipsis site, giving rise to a mood mismatch with its antecedent, while in (25b), the indicative feature is encoded in T, so mood mismatches are immaterial.

To summarize, we need a different way of looking at the possibility of embedded gapping in Spanish, since, unlike in English, it does not hinge on the presence or absence of the complementizer. In the next section, I provide a phase-based analysis of embedded gapping along the lines of Wurmbrand’s (2017) claim that the possibility of embedded stripping is related to the phasehood of the embedded CP.

4 Embedded gapping in terms of phasehood inheritance

Previous work on phase theory has claimed that, contrary to Chomsky’s (2000; 2001) original conception that only vP and CP are phases, phasal domains are not fixed and can be varied depending on syntactic contexts (see, e.g., Den Dikken 2007; Gallego 2007; 2010; Gallego & Uriagereka 2007; Alexiadou, Anagnostopoulou & Wurmbrand 2014). In particular, Chomsky (2015) proposes that downward inheritance of phasehood is possible and provides a phase-based account for that-trace effects. He claims that if C is deleted at the phase level, phasehood of C is inherited by T; T becomes a derived phase, and the complement of T is transferred for labeling purposes. Citko (2020) also discusses another case of downward inheritance of phasehood to account for the difference between English and Polish with respect to the Sluicing-COMP generalization.⁷ The mechanism of phasehood inheritance proposed in Citko (2020) is given in (33), where the Foc head below CP inherits phasehood from C, and the target of ellipsis is the complement of the derived phase, leaving the complementizer intact:

Citko (2020: 47) claims that phasehood inheritance is an extreme case of feature inheritance, and if the original phase head does not have any uninterpretable features left, it does not have to keep on being a phase head (see also Gallego 2010 for a similar claim). In (33), the uninterpretable wh-feature on C is inherited by Foc; thus, CP is no longer a phase.⁸ For shifted phasehood, Citko (2020) provides evidence from sluicing in languages like Hungarian and Polish, where the wh-remnant moves to [Spec, FocP], and claims that in such cases, Foc is the phase head that licenses TP-ellipsis, despite the presence of an overt complementizer in C (see (20)). In this section, I adopt this proposal and apply it to embedded gapping, assuming that ellipsis targets the complement of a phase head (Van Craenenbroeck 2004; 2010; Gengel 2009; Rouveret 2012), and that the Phase Impenetrability Condition (PIC) renders spell-out domains inaccessible to further operations (Chomsky 2000; 2001).⁹ I argue that spell-out domains for ellipsis may vary depending on phasehood inheritance, and this accounts for crosslinguistic variation: English and Spanish differ from each other in terms of the spell-out domains when gapping occurs in embedded clauses. Specifically, English disallows embedded gapping, (34a), since CP is a phase, and therefore the spell-out domain is TopP rather than TP, as in (34b):¹⁰

In this structure, TopP is what should be elided in line with the approach that phase heads trigger ellipsis of their complements. Thus, the ungrammaticality of (34a) is due to the fact that the ellipsis site does not match the spell-out domain.¹¹

In contrast, Spanish allows embedded gapping, (35a), in which case a lower head below CP constitutes a phase; phasehood of C is inherited by Foc, thereby rendering TP the spell-out domain, as in (35b):

In (35b), I assume that C does not transfer its phasehood directly to Foc; the phasal status of FocP is attributed to Top which inherits phasehood from C. Since gapping involves a contrastive topic-contrastive focus structure in which Foc is the licensor of TP-ellipsis (Gengel 2013; Winkler 2015), phasehood inheritance cannot stop at Top; it must go all the way down to Foc.

Under the current proposal, the phasehood of the embedded CP is what determines the possibility of embedded gapping; in Spanish, FocP is a phase that allows the ellipsis of its complement (i.e., TP), whereas in English, TP is not the target of ellipsis, since CP is a phase. The question, then, is why in Spanish, Foc can inherit phasehood from C. Following Van Craenenbroeck & Lipták (2006; 2013), I argue that the availability of wh-movement to [Spec, FocP] is related to the phasal status of FocP. In fact, Spanish is similar to Hungarian in that the [E]-feature is linked to the operator feature in Foc rather than the wh-feature in C; in indirect questions where the wh-element follows the complementizer que, the former moves to [Spec, FocP], while the latter is the head of ForceP of the split-CP structure (see Demonte & Fernández-Soriano 2009). Wurmbrand (2017: 361) claims that the distribution of complementizers in Spanish provides initial evidence for the phasal status of a lower head below CP; in double-complementizer constructions like (36), the locality effect with the doubled que (in boldface) associated with TopP or FocP is expected if FocP is a phase:

Since left-dislocated elements are topics (López 1999), in (36a–b) the relevant head that licenses TP-ellipsis is Top, irrespective of whether it is null or lexicalized as que (see Villa-García 2012; 2015). This would mean that TopP constitutes a phase that allows its complement (i.e., TP) to be elided under the phase-based approach to embedded gapping and stripping.¹²

Given the analysis in (34b), the question now is how the present analysis accounts for the fact that in English, embedded gapping is possible when the complementizer is absent (see (21)–(22)). One possible explanation for this is that the acceptability of embedded gapping in such cases is related to the absence of the complementizer.

Chomsky (2015) claims that example (37a) is grammatical because C-deletion transfers its phasehood to T, rendering TP labeled, as in (37c), whereas example (37b) is ungrammatical because of the failure of labeling; English T is too weak to label and traces are ignored for the purpose of labeling, so TP cannot be labeled if the subject wh-phrase moves from [Spec, TP] to [Spec, CP]:

Under Chomsky’s (2015) dephasing of C whereby T inherits all the features of C, including phasehood, what is crucial is that the embedded CP loses its phasehood by deleting that.¹³ I propose that the ungrammaticality of (38) is similarly attributed to phasehood inheritance as a consequence of dephasing of the original phase head C; if phasehood is inherited by Top and subsequently by Foc, TP is the spell-out domain to be elided, as in (39):

I assume that the matrix verb takes a CP complement with a null complementizer (Pesetsky & Torrego 2001; 2004; 2007; Bošković & Lasnik 2003) and thus depart from Wurmbrand’s (2017) claim that the matrix verb takes a CP-less clauses as its complement in embedded stripping without complementizers (cf. (18b)). Given the analysis in (39), the acceptability of (38) is not due to the lack of a CP-layer in embedded clauses; rather, the absence of the complementizer comes as a result of C-deletion which makes a lower head inherit phasehood, similar to Chomsky’s (2015) account of the lack of the that-trace effect, given that phases are relevant domains for syntactic operations as well as PF-processes.

One may also claim that the contrast between (34) and (38) is related to the truncated structure of CP (Rizzi 2006; 2010; Haegeman 2006; Rizzi & Shlonsky 2007; Shlonsky 2014). Rizzi & Shlonsky (2007) propose that Criterial Freezing and the Subject Criterion are crucial ingredients for the left-periphery of the sentence. The former is a structural requirement on subjects; an element satisfying a criterion is frozen in place, and the latter implies that once moved an element to [Spec, Subj(ect)P], it cannot undergo further movement (Rizzi 2006; 2010). The gist of this proposal is that subjects move to criterial positions in which they are frozen. According to Rizzi & Shlonsky’s (2007) analysis of truncation of CP-structure, an embedded clause containing the complementizer that involves the structure in which that undergoes head movement from Fin to Force under Rizzi’s (1997) split CP-structure, since it conveys both finiteness and declarative force. In contrast, an embedded clause without a(n overt) complementizer involves a truncated structure which lacks FinP and ForceP. If there are no such layers, the subject does not have to move to [Spec, SubjP]; Criterial Freezing does not apply in such cases. Evidence for truncation in cases of complementizer deletion in English comes from that fact that the deletion of that is incompatible with topicalization or focalization:

If this is correct, the contrast between (34) and (38) could be attributable to two different structures of CPs; full-fledged CP vs. truncated CP:

However, the absence of FinP and ForceP does not expect that embedded gapping is compatible with topicalization even when there is an overt complementizer. This is shown by the fact that in embedded gapping, the presence of that yields ungrammaticality:

In view of this, Rizzi & Shlonsky’s (2007) proposal does not help much to find out why the presence or absence of the complementizer has an effect on the acceptability of embedded gapping in English, nor provides any clue for the difference between English and Spanish with respect to embedded gapping. Alternatively, one might consider that in English, there is no embedded gapping per se, and suggest that the string containing the subject and the verb in the leftmost position of the second conjunct is an instance of epistemic/evidential parenthetical, thus explaining the contrast between (43a) and (43b–c) (see Rooryck 2001 for discussion of parenthetical clauses as evidential modifiers):

While it is beyond the scope of this paper to fully explore the parenthetical account of embedded gapping in English, it is noteworthy that while epistemic/evidential parentheticals typically have a first-person subject, there is no such restriction in (43a):

This is much the same as embedded fragments:

Weir (2014) notes that in (45), the matrix subject, if it differs from the first person, must have the rise-fall-rise contour of contrastive topic, and claims that these examples involve true syntactic embedding and do not pattern with parentheticals. The same holds true for (44). Though I leave a detailed investigation of factivity and embedded gapping for future work, I follow Wurmbrand (2017) in claiming that embedded gapping in English can be treated in the same way as embedded stripping (see Wurmbrand 2017: 344 for arguments against the parenthetical analysis of embedded stripping in English).

To summarize, under the current proposal, embedded gapping is analyzed in the light of the phasehood of the embedded CP, which may vary across languages. The analysis proposed here provides a straightforward account of the Embedded Gapping Generalization:

In the next section, I explore the consequences and challenges of the current proposal.

5 Implications of phasehood inheritance in embedded gapping and (pseudo)stripping

As discussed in Section 3, Spanish allows embedded pseudostripping in which the polarity/focus particle follows the remnant:

This contrasts sharply with embedded stripping, which is disallowed in Spanish:

Thus, it is fair to say that pseudostripping, in contrast to stripping, behaves like gapping with respect to the No Embedding Constraint. In this section I argue that the generalization in (46) can be extended to pseudostripping, and discuss some issues that arise from phasehood inheritance in embedded clauses.

Previous research has shown that not only focus particles such as también ‘too’ and tampoco ‘neither’ but also polarity particles like sí ‘yes’ or no ‘no’ can license TP-ellipsis in Spanish (Bosque 1984; Brucart 1987; 1999; Laka 1990; López 1999; Depiante 2000; 2004a, b; López & Winkler 2003; Vicente 2006; Saab 2008). In particular, Depiante (2004b) and Saab (2008) claim that pseudostripping involves an underlying CLLD structure where the remnant is analyzed as a CLLD-ed element in TopP.¹⁴ Under this analysis, pseudostripping (49a) is derived along the lines of (49b), where TP-ellipsis is licensed by the polarity particle in ΣP, a projection proposed by Laka (1990) to host polarity elements:

If this is so, Σ is similarly the phase head that allows TP to be elided in embedded pseudostripping. However, this goes against Wurmbrand’s (2017) claim that FocP is a phase in languages like Hungarian and Spanish where embedded (pseudo)stripping is allowed. As far as the current proposal goes, there are in principle two functional categories (i.e., Top and Σ) eligible to be a phase head in (47a) and (47b):

The question here is which head should be the licensor of TP-ellipsis in embedded pseudostripping. One possibility is that Σ is not a phase head when it merges with TP, and the polarity particle no undergoes head movement to the phase head Top; the ‘Top-no’ complex head is what licenses ellipsis of its complement, as in (51):

This is reminiscent of X-stranding XP-ellipsis, since the polarity particle has moved out of the ellipsis site (i.e., ΣP). However, this does not follow Lasnik’s (1999: 158) generalization; XP-ellipsis is prohibited if XP has lost its head. Lipták & Saab (2014) argue that, for a language to display headless XP-ellipsis, it must independently show both X-movement and XP-ellipsis. However, this claim is not entirely correct and has been refuted by Sailor (2018) showing that in Mainland Scandinavian, VP-ellipsis is possible even though there is no V-to-T movement. Leaving aside the issue of Σ-movement, it is not clear whether the structure in (51) is suitable for Spanish due to the lack of evidence that ΣP rather than TP is elided, given that polarity particles such as sí and no are related to the propositional content (i.e., TP).

The phase-based analysis proposed in the previous section claims that phasehood inheritance allows for gapping in embedded clauses. I extend this analysis to (47a) and (47b): phasehood of Top is lowered onto Σ. Assuming that the remnant of pseudostripping is located in TopP (Depiante 2004b; Saab 2008; 2010), I propose that Top, like Foc, inherits phasehood from C in embedded pseudostripping. This is not much different from the analysis of embedded gapping in the sense that phasehood is inherited from a phase head to a non-phase head. Specifically, Σ is not originally a phase head; it inherits phasehood from Top, and therefore the complement of Σ is the spell-out domain to be elided. The derivation of (47a) is given in (52):

The same applies to (47b) without further discussion. At this point, the question arises as to why stripping in Spanish cannot occur in embedded clauses (see (48)). According to the current proposal, in Spanish, C transfers its phasehood to a lower head (e.g., Foc), so licensing embedded gapping is a particular instance of TP-ellipsis. In a similar vein, the examples in (48) should be grammatical, contrary to fact. I suggest that the reason for this is that only Top and Foc, which are related to the information-structural properties of the remnants, can be the target of phasehood inheritance in embedded (pseudo)stripping. Following Vicente (2006), I assume that the remnant of pseudostripping is in TopP, whereas the remnant of stripping is in FocP:

In embedded pseudostripping, the phasal status of ΣP is attributed to the derived phase head Top; it is by virtue of Top that phasehood of C is transferred to Σ, as in (54a). In embedded stripping, by contrast, phasehood of C cannot be transferred to Foc, since Σ does not inherit phasehood from C, as in (54b):

As an anonymous reviewer points out, pseudostripping is a type of non-local surface anaphora, similar to VP-ellipsis in English (Depiante 2000; 2004b; Saab 2008); thus, the remnant and its antecedent need not be clause-mates, in contrast to stripping (see Brucart 1987 for a similar claim for gapping in Spanish). Note that pseudostripping behaves differently from stripping with respect to backward ellipsis; the former can operate backwards, (55a), like VP-ellipsis, (56), whereas the latter cannot, (55b) (see Brucart 1999):

This may lend support to the claim that pseudostripping does not obey the clause-mate condition. If this is so, then the similarity between gapping and pseudostripping with respect to the No Embedding Constraint would be superficial. I leave this issue for future research.

The analysis proposed above also provides an account of why embedded not-stripping is impossible in English. Merchant (2003) suggests that the licensor of TP-ellipsis in matrix not-stripping is Foc rather than the negation in [Spec, NegP] (Haegeman 1995) or adjoined to ΣP (Ladusaw 1992). In this case, the negative marker not forms a constituent negation with the remnant in FocP (Depiante 2000; Merchant 2003). Assuming this to be the case, I suggest that in embedded not-stripping, the ellipsis site is not TP but FocP, which is the complement of the phase head C:

As discussed in Section 4, Chomsky’s (2015) phase-based account of the absence of the that-trace effect suggests that the deletion of C transfers its phasehood to T. Similarly, if C is deleted in embedded not-stripping, Foc becomes a derived phase; in this case, Foc inherits phasehood directly from C, since there is no ΣP projection. This is illustrated in (58), which is not much different from embedded gapping when the complementizer is absent (cf. (42)).

If this line of reasoning is correct, we would expect examples like (59) to be acceptable:

For most speakers consulted, embedded not-stripping is acceptable if there is no overt complementizer. This would indicate that the deletion of the complementizer plays a crucial role in the acceptability of embedded not-stripping, just as in the case of embedded stripping discussed in Wurmbrand (2017) (see also Section 3).

Under the current proposal, it is also expected that if the embedded CP loses its phasehood, and unless there is a new phase edge for intermediate movement, long-distance extraction will be banned due to the PIC which forces movement to proceed successive-cyclically through phase edges, as schematized in (60a–b):

To see this, consider the following examples, where one of the remnants is a wh-element:

Although the data look like what Ortega-Santos, Yoshida & Nakao (2014) call sluice-stripping, I assume that (61a–b) involve gapping in wh-questions. Note that similar examples like (62) cannot be viewed as sluice-stripping, since they contain two non-wh-remnants (along with a wh-remnant), which does not fit with the term stripping:

These examples are better suited to gapping as far as the number of the non-wh-remnants is concerned; gapping with three remnants is allowed in Spanish (Brucart 1987), though English does not easily allow three remnants if all of these are arguments of the elided verb (see, e.g., Jackendoff 1971: 25). Given that both conjuncts in (61) and (62) are interpreted as two independent wh-questions, they involve their own wh-movement to [Spec, CP] (see Corver & van Koppen 2018 for similar data from Dutch). In these cases, gapping cannot be analyzed under low coordination and must be derived through TP-ellipsis. Under the phase-based analysis of embedded gapping, I argue that (61b) is unacceptable due to long-distance wh-movement in violation of the PIC; Foc inherits phasehood from C, and the Spec of the embedded CP is no longer the phase edge. Thus, in (61b), the wh-element would move directly from the embedded [Comp, VP] to the matrix [Spec, CP], skipping the embedded [Spec, CP]. However, this movement is not successive cyclic, violating the PIC:

In (63), when TP merges with Foc, it is transferred to PF, and further movement to [Spec, CP] is not available due to cyclic transfer (e.g., Uriagereka 1999). Note that the non-elliptical counterpart of (61b) does not violate the PIC, since long-distance wh-movement passes through the edge of a phase (i.e., the Spec of the embedded CP) on its way to the matrix clause:¹⁵

Given this, the grammaticality of (61b) needs to be accounted for under the current proposal. I suggest that in (61b), the wh-element undergoes movement to [Spec, FocP], the edge of the derived phase, and then moves further to [Spec, CP] of the matrix clause:

That is, long-distance extraction is possible out of embedded gapping even though the embedded CP loses its phasehood; [Spec, FocP] serves as a new phase edge for intermediate movement.¹⁶ The question then arises as to whether English also behaves similarly to Spanish in this respect. All else being equal, long-distance extraction should be possible in English, passing through [Spec, CP], but the result is ungrammatical. The reason is that the embedded CP is a phase, and the spell-out domain (i.e., FocP) does not match the ellipsis site (i.e., TP), as argued in Section 4. Given the account proposed here, the derivations of (61b) and (64) are not the same; while long-distance wh-movement is possible in both cases, the path of successive-cyclic movement is different depending on its phasal domain, as schematized in (66):

6 Conclusion

The existence of embedded gapping in languages like Farsi and Spanish poses some challenges to the crosslinguistic validity of the No Embedding Constraint: why is gapping subject to this constraint in some languages but not in others, and if so, how can embedded gapping be accounted for in a principled way? Recent research suggests that the possibility of embedded stripping is related to the phasal status of embedded CP (Wurmbrand 2017), coupled with phasehood inheritance in these contexts (Citko 2020). If gapping and stripping are similar in terms of their restriction to coordinate structures, the difference being the number of the remnants, the same structural constraint for embedded stripping may also apply to embedded gapping and lead to the same empirical consequences.

In this paper, I have argued that the possibility of embedded gapping is determined by whether the embedded CP is a phase or not, on the assumption that the ellipsis site corresponds to a phasal complement. The difference between English and Spanish with respect to embedded gapping lies in the phasal status of the embedded CP. English does not allow gapping in embedded clauses with overt complementizers, since CP is a phase, and therefore the target of ellipsis is TopP rather than TP. However, embedded gapping is allowed in English when the complementizer is absent; if CP loses its phasehood by deleting the complementizer, FocP below CP becomes a derived phase, and TP is the target of ellipsis. In contrast, Spanish allows embedded gapping, since FocP constitutes a phase after phasehood inheritance, rendering TP the ellipsis site. The current proposal shed new light on the correlation between the phasehood of the embedded CP and the possibility of embedded gapping, in particular how the No Embedded Constraint can be circumvented in Spanish, and even in English when the complementizer is absent. I have extended the proposal to embedded pseudostripping in which Σ inherits phasehood from C by virtue of Top, and the complement of Σ is the target of ellipsis. Finally, I have shown that long-distance wh-movement is possible out of embedded gapping when FocP is a phase, in which case [Spec, FocP] is a new phase edge for intermediate movement on its ways to the matrix [Spec, CP], yielding successive-cyclic movement.

Although this paper attempts to show how the phasehood-inheritance-based proposals provide a plausible account of embedded gapping in Spanish and other languages, there are still open questions regarding the deeper nature of the variation in the patterns of phasehood inheritance across different languages, in particular in relation to overt and null complementizers, as well as the phasal status of FocP (or TopP). I leave these issues for future research.

Abbreviations

1 = first person, 2 = second person, 3 = third person, cl = clitic, acc = accusative, dat = dative, dom = differential object marking, impfv = imperfective, ind = indicative, subj =subjunctive, F = feminine, M = masculine, sg = singular, pl = plural

Competing interests

The author has no competing interests to declare.