2.1 Definitions

Head-stranding ellipsis describes an operation in which the head of a phrase undergoes movement out of that phrase, and the phrase itself is then elided, stranding the head. The present article is particularly concerned with ellipses in which a verb is the stranded head. The following naturally-occurring example from Hindi-Urdu features verb-stranding ellipsis as well as subject pronominal drop, meaning that the entire clause of the audience’s response is reduced to the verb alone.

Of course, the systematic study of the syntactic and semantic properties of verb phrase ellipsis (VPE) itself has a long and rich history (Ross 1969; Grinder and Postal 1971; Hankamer and Sag 1976; Williams 1977; and so on), but in-depth research on verb-stranding ellipses (and in particular verb-stranding VPE) did not begin in earnest in the modern literature until about 30-40 years ago in work such as that of McCloskey (1991), Otani and Whitman (1991), Ngonyani (1997), Doron (1999), Sherman (Ussishkin) (1998), Martins (1994), and Goldberg (2005). In verb-stranding VPE, the verb moves out of the verb phrase to a higher head, and then the verb phrase is elided (including all remaining VP-internal material), stranding the verb itself.

As mentioned above, verb-stranding VPE has been argued to occur in a wide range of unrelated and typologically diverse languages including Irish (McCloskey 1991), Hebrew (Doron 1999; Goldberg 2005), Portuguese (Martins 1994), Russian (Gribanova 2013a; b), Hindi-Urdu (Manetta 2018), and Uzbek (Gribanova 2020), among many others. Consider the following examples from Russian and Portuguese:

The now-familiar challenge to unambiguously identifying instances of verb-stranding VPE is that many languages that feature a verb-stranding string that could potentially be generated via VPE also permit DP arguments to go missing in other ways. For instance, the language may allow null pronominals for certain types of objects, or might feature Argument Ellipsis (AE), eliding individual arguments of the verb (Oku 1998; Kim 1999; Takahashi 2008).

Verb-stranding ellipsis could also be revealed to be ellipsis of a clause-sized constituent (TP) out of which the verb has moved.

2.2 Isolating verb-stranding ellipses: a short history and the state of the art

Due to the ambiguity inherent in determining the derivation of the strings produced by the processes in (2), (5), and (6), among others, a battery of syntactic diagnostics have been developed in the last several decades to try to definitively identify instances of verb-stranding VPE/TPE and to differentiate these operations from others that generate similar strings. The strongest of these diagnostics has the capacity to demonstrate that even in a language in which other processes may result in verb-stranding strings, verb-stranding VPE/TPE is necessarily available. Arguments in support of the existence of V-stranding VPE/TPE in a given language generally look like some or all of the following:

• These languages have independent evidence for verb movement out of the VP/vP in non-elliptical structures

• VP/TP-internal material that has gone missing is recoverable in the ellipsis site – in particular material that cannot be independently dropped but can only go missing as part of a larger constituent

• There may be matching requirements placed on the verb in the antecedent clause and the correlate clause (the Verbal Identity Condition (Goldberg 2005))

In her dissertation, Goldberg (2005) offers a review of diagnostics used to identify verb-stranding VPE to that point, including tests which link the characteristics of English-style VPE (which strands an auxiliary verb) to verb-stranding VPE in languages like Hebrew and Irish. Goldberg takes a critical stance toward earlier work which failed to fully recognize the potential for arguments to go missing independently from VPE. She attempts to isolate certain VP-internal constituents, such as manner adverbial PPs and benefactive PPs, and claims that they cannot be elided independently. These diagnostics later prove to be problematic for other languages (e.g. Russian in Gribanova 2013 a;b) and even for Hebrew as discussed in Landau 2018.

At the same time, other researchers have continued to develop new diagnostics (e.g. Simpson, Choudhury, and Menon 2013; Gribanova 2013a;b; Gribanova and Manetta 2018; Gribanova 2020) or investigate the interaction of existing diagnostics with other operations in the grammar (Merchant 2018; Gribanova and Mikkelsen 2018; Manetta 2018; 2020). For instance, Gribanova (2013a) proposes a new diagnostic to isolate verb-stranding VPE in Russian which features ellipsis of a VP containing two coordinated or disjoined VPs (themselves potentially containing multiple elements). Gribanova argues that because the coordinator cannot be independently dropped in this (or any other known) language, when such a complex VP goes missing in its entirety this can only be due to verb-stranding ellipsis. In this fashion, this particular approach to better understanding head-stranding ellipsis has indeed become a highly productive domain of crosslinguistic research and discovery.

Most recently, Landau (2018; 2020a;b) has forcefully argued that a number of the (often longstanding) diagnostics designed to isolate verb-stranding ellipses are ineffective and fail in some or all languages under investigation. For instance, he reviews and ultimately rejects the results of tests for island-sensitivity for object gaps without a linguistic antecedent (first proposed in Doron 1999) and for adjunct inclusion in the ellipsis site (proposed in Goldberg 2005). His explanation as to why these tests fail is not that there are other interfering language-specific or more general factors (see also Bailyn 2014; Erteschik-Shir, Ibn-Bari, and Taube 2013), but that the conditions for head-stranding ellipsis are tightly constrained, and indeed verb-stranding VPE does not exist at all (it is thus unsurprising that it cannot be positively identified).

As it turns out, applying these syntactic diagnostics is remarkably delicate work. Throughout the literature, judgements of native speakers as reported by researchers and even by native-speaker linguists reveal enough variation in acceptability of test sentences that we should be quite cautious in our interpretation of the results. For example, on the topic of the interpretation of adjuncts in verb-stranding VP ellipsis sites in Persian, Rasekhi (2016) claims that the elided adjunct reading is not available in downward entailing environments, though in a footnote (ftnt 7) admits that some speakers can obtain these readings with very strong contrastive stress on the adverb. On the other hand, Toosarvandani (2019) states the elided adjunct reading is indeed available in these environments in Persian without any further discussion (Toosarvandani 2019: 953). Additional instances of this kind of disagreement concerning acceptability judgements and interpretations in verb-stranding environments can be found in Japanese (Funakoshi 2016); Russian (Bailyn 2011; Erteschik-Shir, Ibn-Bari, and Taube 2013); Uzbek (Gribanova 2020); Hindi-Urdu (Manetta 2020a); and Portuguese (Landau 2020a), among others.

It also becomes increasingly clear that when multiple sizes of constituents may be elided, resulting in effectively identical strings, pragmatic, prosodic, and information-structural factors may be at work conditioning the way in which speakers parse structures (Merchant 2018; Manetta 2018; Gribanova, 2020). The diversity and degree of impact of these factors seems as yet poorly understood (though see Simpson (2021) for some recent thinking on which factors might be relevant).

Given these uncertainties, the crosslinguistic empirical picture concerning diagnostics designed to positively identify head-stranding ellipses is far from straightforward. For this reason, the present paper seeks an additional pathway by which to evaluate the newly proposed phase-based account for licensing head-stranding ellipsis. This is not to say, of course, that the careful analysis needed to probe for and isolate verb-stranding ellipses is infeasible; this is critical work pursued in the present article. But in addition to this effort, the investigation that follows explores the potential implications of the phase-based account of head-stranding ellipsis for our understanding of head movement, and in particular the head movement that generates verb-second word orders. If the phase-based account requires approaches inconsistent with what we already know about V2, and we elect to maintain the dominant account of ellipsis licensing at the conclusion of this investigation, note that we are still left with many of the challenges presented by the empirical landscape described above. An important impact of Landau’s recent work is to highlight that those challenges will require our serious attention.

3.1 Word order

Kashmiri is an Indic language, and as such it is relatively unusual in that it requires verb-second word order in nearly all finite tensed clauses, both matrix and embedded.

When a tensed auxiliary is present, the auxiliary takes second position and the main verb appears in a participial form typically either immediately following the auxiliary or in final position (though it can be scrambled elsewhere) (Wali and Koul 1997: 220).

Kashmiri is understood to be underlyingly verb final, given that: (a) the untensed verb appears in clause final position in unmarked non-finite environments, (b) the participial form of the main verb appears in final position in the presence of a tensed auxiliary, and (c) in certain tensed clauses (e.g. relative clauses, comparative clauses) the unmarked position for the verb is clause-final as in (10).

Kashmiri features obligatory wh-fronting to the immediately preverbal position. In wh-questions it is also possible (and for many speakers, preferable) for an additional XP typically interpreted as a topic to appear before the wh-phase, throwing the verb into third position.

The morphology of the tensed verb has been taken to support the view that the verb undergoes syntactic movement to and through T (Munshi and Bhatt 2009). The verbal stem is suffixed with morphemes marking tense and agreement (including a system of agreement clitics conventionally glossed as person suffixes or ps), negation when present, followed by the polar question marker when present (this inseparable sequence is bolded in (12)).

Bhatt (1999) reviews in detail other classic word-order-based evidence for the syntactic height of the Kashmiri verb (Pollock 1989). For instance, lower adverbs follow the tensed second-position verb, but precede the clause-final verb:

Further, though negation is strictly a verbal suffix in V2 environments, in non-V2 tensed relative clauses it appears either immediately following the relative phrase or suffixed to the clause final tensed verb. If the latter, it may not be separated from that verb by other material (even if the verb subsequently scrambles elsewhere). Munshi and Bhatt argue that this data indicates that the tensed verb thus moves to a T in all tensed clauses (whether V2 or not) and then continues into the right-headed Neg⁰ head of NegP in the presence of negation.

The facts above, alongside other word order facts reported in detail in previous work (Bhatt 1999; Munshi and Bhatt 2009) have prompted all recent accounts of Kashmiri V2 to posit that the second position verb (whether main or auxiliary) moves through the tense head T (right-headed) and then continues to a higher left-headed functional head which is identified either as in the C domain (Manetta 2011; 2020) as in (16) below, or between T and C (to a head labeled Mood) (Bhatt 1999; Munshi and Bhatt 2009). For concreteness, I will assume the account of V2 in Kashmiri along the lines of (16), though we will return to discussion of the version proposed in Munshi and Bhatt in what follows.

This account is largely consistent with the standard approach to V2 crosslinguistically, developed with reference to better-studied continental Germanic and Scandinavian languages (e.g. Den Besten 1983; Platzack 1986; Holmberg 1986; Holmberg and Platzack 1995; Vikner 1995; Gribanova and Mikkelsen 2018; Harizanov and Gribanova 2019).

3.2 V-stranding ellipsis in Kashmiri

Kashmiri exhibits what appears to be V-stranding ellipsis, stranding both auxiliary verbs as in (17b), and main verbs as in (18b). In this subsection, if not otherwise cited, acceptability judgements and interpretations for the Kashmiri language data were confirmed in consultation with 16 native speakers. Any disagreement amongst the speakers is noted.

In the case of the string which strands an auxiliary verb as in (17b), there is no question that ellipsis of some larger constituent is involved, as the participial form of the main verb has gone missing alongside the internal arguments but is obligatorily interpreted in the ellipsis site. As mentioned in section 3.1, the tensed auxiliary verb must occupy second position, and is thus understood to be in C. The elided phrase in (17b) could be a verbal projection (vP), or something larger (AuxP or TP) out of which the auxiliary verb has raised for the purposes of generating V2 word order.

Despite the superficial similarity between (17b) and (18b), the mechanisms by which (18b) is generated are less clear. There is no doubt, as in the case of the auxiliary verb, that the main verb in (18b) is in second position, and thus has moved outside the vP in which it originated. However, there are at least four ways in which the Kashmiri lexical items translated as ‘him’, ‘a book’, and ‘tomorrow’ could have been rendered missing.

1. ‘him’ and ‘a book’ are null pronominals; ‘tomorrow’ is accommodated pragmatically

2. ‘him’ and ‘a book’ are elided separately in an instance of argument ellipsis (AE); ‘tomorrow’ is accommodated pragmatically, or undergoes adjunct ellipsis

3. verb-stranding VPE, eliding all VP-internal material

4. verb-stranding TP-ellipsis, eliding all clause-internal material

Analysts typically provide evidence that not all missing internal arguments can be amenable to the null pronominal explanation in (a), by showing that verb-stranding strings admit interpretations unavailable to overt pronominals. For instance, Landau (2018) illustrates the fact that in Hebrew an internal argument consisting of disjoined nominals can antecede a verb-stranding string but not a pronoun. The same holds for Kashmiri:

Further, as in other languages, indefinite readings are not available for pro-dropped arguments (Holmberg 2016; Sato and Hayashi 2018; Gribanova 2020), such as the dropped subject in (20a) below, but an indefinite reading is certainly available for subjects that go missing in verb-stranding contexts with linguistic antecedents, as in (20b).

The data in (19)–(20) confirm that ellipsis is certainly available in Kashmiri (that is, not all missing material be understood to be a null pronominal). The question then becomes: what size of constituent is being elided?

Two very recent diagnostics have emerged in an effort to differentiate strings generated by VP or TP-ellipsis from those potentially generated by multiple instances of AE. The first of these tests, developed in Gribanova (2013a), shows that coordinated or disjoined verb phrases can be elided as a unit, stranding a common verb. While AE could be the mechanism eliding the internal arguments, there is no known mechanism that allows a coordinator to go missing independently.

A second diagnostic, developed in Gribanova (2020), depends on the claim that AE can elide arguments, but not predicates. To establish this clearly, we can show that there are predicate constituents that cannot go missing on their own but can quite naturally go missing as part of a larger elided constituent. The sentences in (22a–b) illustrate by interpretation that resultative predicates of verbs and adjectival phrase predicates of verbs cannot go missing independently. By contrast, the data in (23a–b) show that these same kinds of predicates can go missing and be necessarily interpreted as included in the ellipsis site if part of a larger missing constituent (preceded by a stranded main verb).

It is the obligatory inclusion of the predicate in the sentences in (23) that reveals the operation of interest to be V-stranding ellipsis.²

The results of these preliminary tests, coupled with the fact that Kashmiri is clearly a language with verb movement to a position well outside of the verb phrase, would suggest that this language exhibits verb-stranding ellipsis of a larger constituent (vP or TP).

Several additional diagnostics which exploit the properties of Kashmiri V2 indicate that the ellipsis is indeed clause-sized (that is, this is V/Aux-stranding TPE). First, since any constituent can occupy first position so long as the verb is in second position, we should expect to see non-subject constituents followed by the verb preceding an ellipsis site, as in (24). In this string we have evidence of extraction of vP-internal constituents out of the ellipsis site, confirming that the missing material does not correspond to a proform. Further, the subject, which by all accounts occupies Spec, TP when it is not preceding the V2 verb, is missing as part of the ellipsis site. Since an indefinite reading is not available for pro-dropped subjects, as discussed in reference to (20) above, we can control for independent subject pro-drop by using an indefinite subject in the diagnostics that follow. The strings in (24) are then best understood as ellipsis of a TP.

The auxiliary-stranding string in (24b) is unquestionably TP-ellipsis, as the participial form of the main verb liichmIts ‘wrote’ goes unpronounced in the second clause. In order to ensure that main verb-stranding strings are also generated by TP-ellipsis and not multiple individual instances of argument ellipsis, we can draw on the diagnostics above in (19)–(23), combining those with scenarios in which the preverbal constituent is a non-subject. In (25)–(26), a non-subject initial constituent precedes the verb, and the interpretation illustrates that a predicate (which we have shown cannot go missing independently) is understood to be obligatorily present in the ellipsis site. In (27b), a non-subject initial constituent precedes the verb, and disjoined constituents go missing.

Sentences like those in (25)–(27) indicate that Kashmiri has a process of main verb/aux-stranding TP-ellipsis. Further new support for this conclusion stems from tag questions.

An old insight that dependent tag questions are generated via ellipsis over an interrogative clause (Huddleston 1970) is further developed in Sailor (2013), who shows that in a range of languages, the set of heads available for stranding in dependent tag questions in a language is precisely the set of heads stranded in regular head-stranding ellipsis structures. For instance, English features ellipsis structures that strand auxiliary verbs but not main verbs (Sita is meeting with the candidate, but Tara isn’t.), and tag questions with auxiliary verbs but not main verbs (Tara left, didn’t she/*left she?). On the other hand, Irish features main verb-stranding ellipsis, and permits main verb tags (McCloskey 1991: 273–74; Sailor 2013).

This parallelism pans out in Kashmiri as well. Kashmiri features head-stranding ellipsis with both auxiliary and main verbs, and both auxiliary and main verbs can be stranded in dependent tag questions (Manetta 2020b):

Expanding the scope of Sailor’s insights slightly (which are restricted to VPE), the claim is that dependent tag questions are generated via whatever head-stranding ellipsis process(es) is/are available in the language applying to a root question, explaining the set of parallelisms.

These facts are of relevance here because they suggest that we could not explain verb-stranding strings in tag questions as generated by only multiple instances of AE. Ellipsis of all VP-internal argument material in dependent tags is obligatory. That is, grammatical tag questions crosslinguistically include only the verbal tag (whatever it may be), polarity particles and/or negation, and in some cases the subject, external argument, or expletive pronominal (Sailor 2013). It is ungrammatical for, say, one of the internal arguments to remain.

Since each separate instance of AE should be optional, and there is no obvious mechanism by which the ellipsis of any one internal argument can be linked to the ellipsis of another, this fact is hard to explain under the AE-approach to head-stranding strings. Consider Portuguese, for which Landau (2020a) claims that for at least some speakers, verb-stranding strings are derived solely by AE.⁴ In both European and Brazilian Portuguese, no internal arguments may grammatically remain in the tag (data courtesy of Ana Maria Martins, p.c.).

Under Landau’s account of main verb-stranding strings in Portuguese — that they are only generated by AE and not by ellipsis of larger constituents — there is no straightforward way to prevent the partial deletion structures in (31b)–(32b). Of course, in Kashmiri it is equally ungrammatical for an internal argument to remain following the main verb in a tag question:

This data supports our position that the AE-only reanalysis of main-verb stranding strings is not viable for languages like Kashmiri and Portuguese, which feature elliptical strings stranding both main verbs and auxiliary verbs (Martins 2016; Manetta 2020b). At the very least, verb-stranding via ellipsis of a larger constituent must be available in some domains in order to correctly generate dependent tag questions (see also Landau, 2020b: 302).

In summary, we have seen a variety of types of empirical evidence which lead us to conclude that Kashmiri features TP-ellipsis stranding both main and auxiliary verbs. We now ask: how would the head movement restrictions imposed by the phase-based account of head-stranding ellipsis licensing (Landau 2020a;b) interact with the verb movement required to create verb-second word orders in a language like Kashmiri? Would it be possible to generate auxiliary and main verb-stranding TP-ellipsis strings in Kashmiri under the phase-based account?

5.1 How high is V2?

As discussed in detail in section 3 above, recent approaches to verb-second in Kashmiri posit that the tensed verb undergoes head movement to C (Manetta 2011, 2020b). Further, standard approaches to V2 assert that auxiliary and main verbs move to the same head for the purposes of creating V2 word orders, and there is no evidence to the contrary for Kashmiri (e.g. den Besten 1980; Vikner 1995). The default assumption would then seem to be that all tensed verbs in Kashmiri move to the C head as in (46) and (47):

In both scenarios, the composite head V-v-T/Aux-T head moves to C.

Turning to verb-stranding and Aux-stranding ellipsis, under the phase-based approach to ellipsis licensing the CHSE prohibits the ellipsis of a category XP if its head X has crossed outside of a spellout domain at the time at which the licensor is introduced into the derivation. Recall that this is a PF condition, in that the host of [E] must be PF-visible in its Spell-Out domain. Under this account, there seems to be no configuration in which headless ellipsis of TP, stranding the tensed, second-position verb (main or auxiliary) is licensable. In either scenario, T would have crossed outside of its spellout domain (TP) to the C head (as part of a composite verbal head) at the point at which the licensor C has been merged. Much like in the case of the blocked Aux-to-C movement prior to TPE in English sluicing (schematized in (39) above), TP ellipsis can never be licensed in this configuration. Thus, given standard assumptions about V2 (that both aux and main verbs move to C), the phase-based account predicts no verb-stranding TP-ellipses to be available at all in Kashmiri, contrary to fact.

A first alternative to reconsider is the height of the tensed verb in V2 structures: what if the verb does not move all the way to C, but instead to some lower head beneath the phase boundary? Indeed, there is a relatively recent alternative approach to Kashmiri V2 which proposes that V2 in the language is head movement of the tensed verb (main or auxiliary) to a Mood head, found above TP but below C (Munshi and Bhatt 2009), as in (48).

Head-stranding TP ellipsis is indeed possible in the configuration in (48). According to the CHSE, TP can be targeted for ellipsis here because the T head (as a part of the composite V-v-T/Aux-T head) has not crossed a phase boundary on the way to the Mood head. In fact, verb-stranding ellipsis as TP-ellipsis with the verb in a head between T and C looks structurally analogous to cases of Polarity ellipsis discussed above, and is licensable in the phase-based view (Landau 2020b). While this might at first seem to suggest that the phase-based account of head-stranding ellipsis can handle the facts of Kashmiri if this modified V2 approach is adopted, by which the second-position verb moves to a head lower in the clausal spine than C (within the same spellout domain as T), a problem arises when we consider sluicing.

5.2 Sluicing and head movement in Kashmiri

It is not unexpected that Kashmiri, a language with obligatory wh-fronting to a position immediately preceding the second position verb, would feature typical sluicing (clause-sized ellipsis, stranding a wh-phrase).

Indeed, languages with similar V2 patterns, such as Icelandic and Yiddish, also exhibit sluicing:

Again, an important accomplishment of the phase-based approach, as discussed above in section 4, is a straightforward account for the well-known fact that the auxiliary verb never appears in sluicing structures in languages like English (a component of the Sluicing-Comp generalization), even when subject-auxiliary inversion should have applied, sending the auxiliary verb to the C head (repeated in (51)).

Returning to Kashmiri, let us imagine that we have adopted the view that both auxiliary and main verbs move to a Mood head, above T but below C. Notice that in that scenario, TP can be elided while the main or auxiliary verb remains present, since neither complex head would have crossed outside the relevant spellout domain. This means that we should expect an overt verb to obligatorily appear in typical sluicing structures as well, contrary to fact, as it would have moved to Mood⁰ to satisfy V2. In essence, in rejecting the account of Kashmiri V2 in which the second position verb moves to C in order to accommodate the phase-based account of verb-stranding ellipsis as discussed in 5.1, we run into the problem of losing the explanation for the missing verb in sluicing contexts (an explanation central to the phase-based account to begin with).⁷

One could propose yet another adjustment in the way in which verb movement functions in Kashmiri, in which in interrogative contexts only, the second-position verb moves to C; otherwise it is found lower in the clausal spine, in the Mood head. However, there are empirical reasons to reject this move. In Kashmiri, a special word order variation is available only in wh-interrogatives in which an additional non-wh DP (with a topic or focus interpretation) may appear preceding the wh-phrase, creating V3 order. Indeed, many speakers report that this word order, in (52), is preferred in unmarked wh-questions (Wali and Koul 1997).

Notably, the V3 word order for interrogatives is also found in Yiddish, suggesting that this is not an exceptional configuration found only in Kashmiri, but part of a wider pattern in a set of V2 languages.

In order to accommodate Kashmiri word orders like those in (52), Munshi and Bhatt (2009) posit that V2 in interrogatives is actually head movement to a head lower than V2 in declaratives (they term this head the Wh/Focus head). In their account, the pre-wh phrase is found in the specifier of MoodP, and the wh-phrase is found in the specifier of Wh/FocusP, with the verb in the Wh/Foc head.

The phase-based account of ellipsis licensing would force us into an approach to V2 in Kashmiri in which we would need to postulate that V2 in interrogative clauses would be movement to a head higher than in non-interrogatives; this is the opposite of what Munshi and Bhatt propose in (54), in which interrogative V2 is to a head which is lower in the clausal spine. This would result in a loss of the explanation for the ordering of preverbal constituents in wh-questions.

Finally, a reviewer asks whether we could consider sluicing to be wh-movement to Spec, CP followed by MoodP ellipsis, with the verb remaining in the Mood head and thus going unpronounced. Although this would derive the desired lone wh-remnant, we could no longer maintain the Ross-Merchant position that the internal syntax of a sluice should be a strategy of wh-question formation independently available in the language. Not only would the adjustment the reviewer proposes not align with a known wh-question formation strategy in Kashmiri, but the special structure needed to feed sluicing would be sharply ungrammatical in the absence of ellipsis — the word order WhXP-Topic-Verb is not possible (Wali and Koul 1997; Manetta 2011).

The upshot of the examination of V2 in this section is that the neither the more widely-accepted view of verb-second, in which both auxiliary and main verbs are found in the C head, nor an alternative view developed in reference to Kashmiri V2, in which tensed verbs are found in the Mood head in non-interrogatives and in a lower Wh/Focus head in interrogatives, are compatible with the requirements of the phase-based account of verb-stranding ellipsis. No empirical evidence supports the approach to V2 imposed by the phase-based approach to ellipsis licensing – indeed, it would seem to be contravened by the data at hand.

In the most general sense, the phase-based account predicts that no verb-stranding ellipsis options would be available in languages with obligatory long-distance head movement to the C domain (V2 languages), but this does not seem to be the case. In section 6 we shall see that the more conventional account of verb-stranding ellipsis along the lines developed in Merchant (2001, 2008) and more recently extended in Gribanova (2013a,b; 2017), has clear advantages in that it works well in concert with a standard, well-supported approach to V2 crosslinguistically.

5.3 Other approaches to V2

In fact, any account of V2 language in which main or auxiliary verbs move directly to a head in the C domain without stopping in the T domain will face incompatibilities generating head-stranding ellipsis strings under the phase-based account. Consider a language like Danish, which permits auxiliary-stranding strings only (Mikkelsen 2006; 2015; Houser et al 2008):

Presumably Landau would analyze Danish, like English, to exhibit Aux-stranding AuxPE.⁸

Several accounts of V2 in Danish propose that V2 is movement of the tensed verb directly to C, with no intermediate landing site (Vikner 1995; Sailor 2018; Harizanov and Gribanova 2019). Consider this approach to V2 under the phase-based account: if the auxiliary verb does not move at all until C is merged, and then goes directly to C, it would be either: (a) still inside the ellipsis site AuxP when T is merged, meaning ellipsis of AuxP would not strand Aux if T is the ellipsis licensor; or (b) moved outside the spellout domain to C at the point at which C is merged, prohibiting headless ellipsis of AuxP by the CHSE if C is the licensor. In either case, licensing Aux-stranding AuxPE would be impossible. Of course, Danish features typical sluicing with no auxiliary verb remaining (Merchant 2003; Chung 2005), so once again, simply positing that Aux-stranding ellipsis is TP-ellipsis licensed by C will not solve the problem. We would then be unable to generate grammatical sluicing structures, and unable to limit verb-stranding ellipsis to auxiliary verbs in this language.

This brief discussion of Danish illustrates that the phase-based approach is not only incompatible with accounts of V2 in Kashmiri, but also with at least some approaches to verb movement in more familiar V2 languages.

5.4 Interim summary

This section has made a number of attempts to extend the phase-based approach to ellipsis licensing to verb-stranding ellipses in Kashmiri and has revealed that the phase-based approach is incompatible with the dominant accounts of V2 in the language (Munshi and Bhatt 2009; Manetta 2011; 2020). This means that no matter where we postulate that main and auxiliary verbs are found in the clausal spine in V2 word orders, the phase-based approach cannot correctly predict the fact that main and auxiliary verb-stranding ellipses are available, alongside typical sluicing structures. Ultimately, the phase-based approach encounters serious challenges in explaining head-stranding ellipsis in a V2 context. For this reason, in what follows I propose that more conventional approach to head-stranding ellipsis following Merchant (2001; 2008) combined with an understanding of V2 as syntactic head movement to C, provides a straightforward account of head-stranding TP-ellipsis in Kashmiri (and likely quite generally).

6.1 A conventional account of head-stranding ellipses in Kashmiri

I maintain here that a relatively conventional approach to head movement and ellipsis in Kashmiri offers a straightforward account of the phenomena discussed above. For completeness, I will illustrate how this account functions before moving on to the larger, more speculative question of how to handle crosslinguistic variation in head-stranding ellipsis.

I assume that V2 in Kashmiri is main and auxiliary verb movement to C as proposed and defended in detail in Manetta (2011; 2020) and consistent with proposals in Vikner (1995) and more recently Harizanov and Gribanova (2019). Following Manetta (2020b), I understand this head movement to take place in the narrow syntax and proceed via head-to-head adjunction through T on the way to the C domain (see also Sailor 2018). I also adopt the dominant approach to ellipsis (Merchant 2001; 2004; 2008), in which a category XP is marked for non-pronunciation when its selecting head Z bears the [E] feature. Head-stranding ellipsis occurs when X⁰ has moved outside of XP and XP (or a containing phrase) is elided, stranding X.

For a typical (discourse-neutral) instance of verb-stranding ellipsis in Kashmiri, I propose that the auxiliary or main verb moves to T, which has a feature to motivate verb movement ([V*]). The T head also has a feature attracting the external argument to its specifier. The complex head V-v-T or Aux-T subsequently moves to C, and the preverbal constituent moves to Spec, CP. C also hosts the [E] feature, prompting ellipsis of TP.

We can now turn to the question of sluicing, which presented an obstacle for the phase based account. Recall that one of the enduring puzzles in the conventional approach to sluicing (Merchant 2001) is the question of why material typically located in C, including complementizers, auxiliary verbs in languages with subject-auxiliary inversion, clitics, and second position verbs, are not present in the sluice (the remnant consists solely of a wh-phrase). This observation was traditionally captured in the Sluicing-Comp generalization (Lobeck 1995; Lasnik 1999; Merchant 2001), which requires that in sluicing structures no non-operator material appear in Comp (Merchant 2001: (71)). In the phase-based account, Landau (2020b) proposes a novel solution to a part of this problem: the CHSE prohibits the movement of a complex verbal head including T up to C and subsequent ellipsis of TP. However, as we saw above in section 5.2, the account of verb-stranding ellipsis in V2 languages forced by the CHSE would predict that the verb remain in sluicing, contrary to fact.

There are a number of viable approaches to the Sluicing-Comp generalization circulating in the literature, including earlier accounts based on the notion of feature strength (Lasnik 2001; Merchant 2001) and a set of accounts relying on a split CP-projection (such that the missing head material appears in a lower head that the one that licenses ellipsis) (e.g. Baltin 2010; Yoshida et al 2015). While space does not permit a detailed review of these approaches here, suffice it to say that many of them could certainly be adapted to work well with the account of ellipsis and V2 in Kashmiri thus far presented. Here I will adopt a mechanism pursued in various forms in Ott (2011), Arano (2014), and Bruening (2015) (see also Wurmbrand 2017, ft 10), which defines the spellout domain targeted for non-pronunciation such that in in certain configurations, the phase head is spelled out together with its complement. In this view, if sluicing is a scenario (marked by a C bearing [Q] and [E]) in which C is included in the spellout domain, and thus goes unpronounced, the second position tensed verb will not appear in sluicing structures.

Very little is new in the conventional approach to Kashmiri proposed in this section. What is important here is to demonstrate that a coherent account of ellipsis already exists that is fully compatible with a well-supported view of verb-second. An account along these lines represents a viable alternative to the phase-based account in Landau (2020b), which requires a series of unsatisfying and unjustifiable moves to accommodate verb-second.

6.2 Future research: crosslinguistic variation in head-stranding ellipsis

One of the most valuable outcomes of Landau’s recent work is to point to a persistent question prompted the ellipsis literature to this point: is there a systematic way to predict the types and sizes of ellipsis that will be available in a given language based on an independent set of characteristics? How can we avoid appealing only to what Landau calls “irreducible, idiosyncratic parametric statements, amounting to a list of the functional heads (or even lexemes), in each language, that license ellipsis of their complements” (Landau2020b: 287). This is certainly not a new question, but it is one newly relevant, as the phase-based licensing condition for head-stranding ellipsis discussed here is but one a series of new efforts to establish the systematic relationship between head movement and ellipsis licensing in the grammar (Schoorlemmer and Temmerman 2012; Thoms 2010; Lipták and Saab 2014; Aelbrecht and Harwood 2015; Sailor 2018; Gribanova 2018, 2020; Bennet, Elfner, and McCloskey 2019; i.a.).

These diverse proposals (often dealing in detail with only a single language) include claims about when head-stranding ellipsis should be available based on head movement patterns, the relative timing of these operations, and the locus of these operations in the grammar. The degree of variation in approaches across these accounts and the empirical challenges they face indicate that there is still real work to do.

First, a complete account would ideally explain why some languages lack certain head-stranding ellipses altogether even when they would otherwise seem to have all of the necessary characteristics; for instance, Landau (2018) argues convincingly that Hebrew lacks verb-stranding VPE, despite the fact that it was long thought to be the prototypical instantiation of the construction. Further, we must ask how extra-syntactic factors such as information structure, argument structure pragmatic cues, and prosody influence the felicity of head-stranding ellipsis in context. Finally, we must better understand how head-stranding elliptical clauses are to be parsed when they are ambiguous with respect to the size of the elided constituent. How does the speaker, the hearer, or, for that matter, the learner interpret such a string in real time, and by what rubrics (e.g. economy) are such strings evaluated? What components of the grammar are engaged in such an evaluation? While these questions lack definitive answers, our understanding of crosslinguistic variation in ellipsis operations and how they interact with head movement will remain partial. I maintain that exploring the empirical landscape of lesser-studied languages like Kashmiri is crucial to testing new hypotheses in this domain.