2.1 Accusative subjects are embedded arguments

A straightforward argument that the accusative DP is an argument of the embedded clause is provided by idiom preservation. Accusative case does not disrupt the idiomatic reading of a subject–verb idiom in the embedded clause (more precisely, the fixed part of the subject in (11) is X’s eyes). If the accusative DP in (11) were a matrix argument, as in the prolepsis analysis (10), the idiomatic reading of the embedded subject–verb idiom could not be accounted for. In contrast, if the accusative DP is the embedded subject, the preservation of the idiomatic reading can be readily accounted for.

Furthermore, in (3), repeated here as (12), the accusative DP follows an unambiguously embedded adverb, suggesting that it is also inside the embedded clause.

Similar data can be seen in (13), where either a nominative or an accusative subject can follow the adverb magadgüi ‘maybe’. (For completeness, (13) also shows different positions that this adverb can occupy.)

But there is an alternative derivation to obtain the same linear order as that in a sentence like (12). We could say that first margaash ‘tomorrow’ scrambles into the matrix clause (i.e. it long distance-scrambles), accompanied by a similar movement of the embedded subject Dulmaa to a matrix position below margaash. The resulting linear order would be the same.

Independently, though, it appears that margaash cannot scramble into the matrix clause (15), regardless of the case of the embedded subject. (More generally, we will see in §3.1 that there seems to be a ban against long distance scrambling in Mongolian.) As such, it is likely that in a sentence like (12), the accusative subject is indeed inside the embedded clause.²³

The claim that accusative DPs are embedded arguments can also be backed up by the fact that the whole embedded clause can be scrambled, while still containing an accusative subject.

Once again, we must rule out an alternative derivation for sentences like (16b). We could in principle interpret (16b) as involving movement of the the possessum sharlovan ‘carrot’ into a matrix accusative case position, before the whole clause remnant-moves to the right of the matrix accusative case position. The superficial result is the same linear order as that seen in (16b).

However, if these movements are effected in a non-string-vacuous way, the result is ungrammatical. In (20), Dorj moves to a position between the matrix subject (Bat) and the matrix adverb changaar ‘loudly’ and the remainder of the embedded clause scrambles to the beginning of the sentence. (We will see in §3.1 that an explanation based on the Proper Binding Condition can be given to (20).)⁴

A fourth argument for the embedded base-generation of accusative DPs is furnished by NPI licensing. (21) and (22) illustrate the basics of NPI licensing in Mongolian: negation is obligatory, regardless of whether the NPI is an object (21) or a subject (22).

In (23), a clause with an object NPI is embedded under the verb khelekh ‘say’. The negation has to be clause-mate with the embedded NPI; matrix negation renders the sentence ungrammatical.

(24) is a similar paradigm, but now the NPI is in the embedded subject position. Once again, negation has to be clause-mate with respect to the NPI; placing negation in the matrix clause renders the sentence ungrammatical. Crucially, the NPI subject can bear nominative or accusative case, but the licensing conditions remain the same. While I cannot offer a theory of NPI licensing in Mongolian, it suffices here that nominative and accusative subjects behave in the same way.⁵

A final argument is provided by the interpretation of nonreferential DPs. Shklovsky & Sudo (2014) observe that, in Uyghur, the subject of an embedded finite clause can be a nonreferential DP that is interpreted under the scope of the matrix verb. This holds regardless of whether the subject bears nominative or accusative case. The same can be seen in Mongolian:

If the accusative DP were a matrix argument, we would not be able to account for the fact that (25) is not contradictory, irrespective of the case of the embedded subject.

We can conclude from the data above that accusative DPs are base-generated as the embedded subject. These data, therefore, argue against the prolepsis analysis in (10).

Nevertheless, nothing was said about the particular position that accusative subjects occupy. It could well be the case that subjects receive accusative case from the matrix v while still occupying the canonical subject position, plausibly Spec-TP. This would be an ECM analysis of accusative subjects in Mongolian, schematized in (26), where the embedded subject receives accusative case without leaving its canonical position, Spec-TP.

In §2.3, I will show that accusative subjects occupy a higher position inside the embedded clause than nominative subjects, leading us to the conclusion that an ECM analysis is empirically incorrect. Before that, I lay out the theoretical problem that the assignment of accusative case across a clausal boundary poses.

2.2 Embedded accusative subjects receive accusative across a clausal boundary

What we have seen so far is that, in Mongolian, accusative DPs that are interpreted as the embedded subject are not proleptic arguments; they are interpreted as the embedded subject by virtue of being base-generated in that position.

The accusative case in the embedded subject has a matrix source. The most obvious argument in support for this claim is that accusative case is not available in root clauses. This can be illustrated with a type of possessive construction where the possessum surfaces without any overt case marking. When embedded however, accusative case becomes a possibility in a possessum.

In view of the unavailability of accusative case in the possessive clause itself (27a), it seems reasonable to assume that this case is coming from the matrix clause in (27b).

Furthermore, accusative case ceases to be an option when the matrix verb is passivized, as in (28), and in subject clauses like those in (29).

The unavailability of accusative case in (28) can be explained if the source of this case is the matrix v, an ability that is taken away as a consequence of passivization. Similarly, in the subject clauses in (29), the subject of these clauses is outside of the c-command domain of the matrix v, hence why it cannot receive accusative case from it.

As emphasized by most literature on similar constructions in other languages, the assignment of accusative case to the subject of an embedded clause across a clausal boundary poses a challenge to common assumptions about locality. In minimalist terms, locality restrictions can be stated in terms of the Phase Impenetrability Condition in (30).⁶

In §3, I will put forward an analysis of accusative subjects in Mongolian that sidesteps the PIC problem by exploring the escape hatch that is hardwired into the definition of this restriction. This proposal is schematized below:

Although both are embedded arguments, accusative and nominative subjects are different in that the former seems to occupy a higher position than the latter. This claim will be based on binding.

2.3 Accusative subjects are higher than nominative subjects

The first suggestion that accusative subjects move to a position that is higher than the position where nominative subjects surface is provided by embedded possessive constructions. (32a) shows that, if the possessum follows the dative possessor, it cannot bear accusative case. (32b) in turn shows that, if the possessum is to the left of the possessor, accusative case becomes a possibility. This is remiscent of Deal’s (2017) observation that a similar construction in Nez Perce may target embedded objects, as long as they are preverbal.

This pattern can be accounted for if accusative case in embedded subjects requires the movement of that DP to a higher position. However, (32b) is still consistent with the ECM analysis in (26), since the accusative DP is just higher than the possessor, but nothing further is said about its position.

A more compelling argument that accusative subjects occupy a higher position than their nominative counterparts is supplied by binding. The reflexive possessive -AA (whose form is subject to vowel harmony) has to be locally bound. In (33a), -AA is attached to the only nominal in the sentence. The sentence is ungrammatical as a consequence of the lack of an antecedent. Indeed, if -AA is taken away, as in (33b), or if an antecedent is provided, as in (33c), the result is grammatical. (33d) shows that the presence of an antecedent is necessary, but not sufficient. In this sentence, -AA is interpreted as the subject of the most deeply embedded nominalized clause. The antecedent has to be the closest subject (Nara) and cannot be the highest subject (Bat).

With this background in place, consider now the behavior of the reflexive possessive -AA in the subject of embedded finite clauses. If it is appended to the nominative subject of an embedded clause (34), the sentence is ungrammatical. Given what we just saw about the licensing of the reflexive anaphor, we could perhaps explain the ungrammaticality of the sentences in (34) as violations of Condition A: the reflexive possessive cannot be bound by the matrix subject because the latter is outside the binding domain of the reflexive. This in turn implies that the embedded clause can be the binding domain of the reflexive possessive within the embedded subject.

Conversely, if the subject is accusative, the result is grammatical. Alternatively put, even though in all the data we have seen so far nominative and accusative subjects can alternate with each other, accusative case becomes obligatory if the Condition A-obeying reflexive possessive is suffixed to the embedded subject. Importantly, in (35a), the accusative subject containing -AA follows the unambiguous embedded adverb margaash ‘tomorrow’, suggesting that the accusative subject is still inside the embedded clause. Likewise, in (35b), the embedded clause as a whole was scrambled, including the -AA-containing accusative subject. Nonetheless, even though the accusative subject remains inside the embedded clause, the reflexive possessive contained in it can be bound by the matrix subject.

If the accusative subject in these sentences were occupying the same position as a nominative subject, (35a) and (35b) should be as ungrammatical as the sentences in (34), contrary to fact. How then can we account for the fact that the accusative subjects in (35a) and (35b) seem to stay inside the embedded clause, while also being bound by the matrix subject? The explanation of the ill-formedness of the sentences where the embedded subject bears nominative case is straightforward: we could take for granted that canonical, nominative subjects occupy a position like Spec-TP and that the binding domain of a DP in this position is restricted to the embedded clause, excluding the matrix subject. (34) would therefore be an instance of a Condition A violation: the matrix subject cannot bind the reflexive possessive because it is outside of its binding domain. The question that we must ask now is which position an accusative subject could be occupying, so that it can be pronounced inside the embedded clause and at the same time allow the matrix subject to bind an anaphor suffixed to it. I will suggest in §3 that a left periphery position, higher than Spec-TP could be appropriate to capture these properties.

Before that, to drive the point home, we will examine similar data, but with the anaphor ööröö. Bondarenko (2017) observes that the distribution of a similar anaphor in Buryat, a Mongolic language that is closely related to Mongolian, argues is favor of a positional distinction between nominative and accusative subjects, the latter being higher than the former. Similar data can be reproduced in Mongolian. Like -AA, ööröö needs to be bound by the closest subject. In (36), the embedded clause is a nominalized clause marked with dative case and the object anaphor there has to be bound by the local subject (pro or Bat). The matrix subject (bi ‘1SG’) is not a viable antecedent.⁷

If we use the genitive form of this anaphor, it can be used as the possessor in a subject. Mirroring the data in (35b), if the subject is nominative, the result is ungrammatical. Again, the ill-formedness of the sentences in (37) can be analyzed as a consequence of a violation of Condition A: the matrix subject is too far away to bind the anaphor, so that it is left antecedent-less in the embedded clause.

In contrast, if the embedded subject containing öör-iin-kh-öö is marked with accusative case, the result is grammatical.

As in (34), if the accusative subject in (38) were occupying exactly the same embedded position as the nominative subject in (37), the former should be as ungrammatical as the latter, contrary to fact. Conversely, if the accusative embedded subject is occupying an embedded position that is higher than the canonical subject position (here, assumed to be Spec-TP) and, concomitantly, accessible to a matrix antecedent, these can be correctly accounted for.

Similar conclusions can be drawn from the occurrence of an anaphor that is itself the subject of the embedded finite clause. In this configuration, the sentence is grammatical only if the anaphoric subject is marked with accusative case:

Again, this nominative vs. accusative contrast can be explained if subjects assigned accusative case occupy a higher position than subjects that are assigned nominative case.

Finally, a Condition B contrast offers another argument that accusative subjects occupy a higher position than nominative subjects. Baker & Vinokurova (2010: 616) show that, if pronominal, accusative subjects in Sakha may not be coindexed with the matrix subject. Nominative pronominal subjects, in contrast, can. Something along these lines hold of Mongolian too, except that Mongolian (unlike Sakha) allows for both nominative and accusative subjects to follow and embedded adverb like margaash ‘tomorrow’. As mentioned above, this helps locate the accusative subject inside the embedded clause.⁹

Yet again, this Condition B contrast can be explained if nominative subjects occupy a position like Spec-TP, which does not include the matrix subject in its binding domain. In this configuration, no Condition B violation is incurred. Accusative subjects, however, would occupy a higher embedded position that does include the matrix subject in its binding domain. As a consequence, Condition B is violated if the matrix subject and the embedded accusative pronominal subject are coindexed.

In sum, it seems that these Condition A and B effects could not be captured if nominative and accusative subjects occupied exactly the same position. The ECM analysis schematized in (26) is therefore inadequate to account for these data. In the next section, I lay out an analysis able to account for these binding effects that is based on a difference in height of nominative and accusative subjects in Mongolian.

3.1 Hyperraising through Spec-CP

Even though accusative subjects in Mongolian can remain inside the embedded clause, they can also be pronounced inside the matrix clause.¹³

(55b) and (56b) show that the accusative subject can also be the leftmost DP in the sentence.

This type of phenomenon was dubbed hyperraising by Ura (1994). It can be found in several other, in some cases unrelated, languages: Buryat (Bondarenko 2017), Korean (Yoon 2004; 2007), Japanese (Bruening 2001; 2002; Tanaka 2002; Takeuchi 2010, i.a.), Nez Perce (Deal 2017), Janitzio P’urhepecha (Zyman 2017; 2018), Sakha (Baker & Vinokurova 2010), Romanian (Alboiu & Hill 2011; 2016), Turkish (Şener 2008), Zulu (Halpert & Zeller 2015), i.a.

We have already discussed prolepsis in §2. This discussion is pertinent again, since the hyperraising variety of accusative subjects in Mongolian is linearly identical to prolepsis constructions (cf. (8)). There are a few argument that suggest that hyperraising in Mongolian is not prolepsis either.

First, if the subject part of a clausal idiom bears accusative case, it can, to a certain degree, be pronounced inside the matrix clause:¹⁴

Second, while embedded proleptic pronouns can occupy different syntactic positions (cf. object position in (8b)), the accusative DP in the Mongolian construction of interest here is interpreted as the subject; it cannot be interpreted as the embedded object:

Third, accusative DPs cannot coexist with an overt pronoun in the embedded clause, as we can see in (59). If the accusative DP and the embedded subject were independent positions (albeit coindexed ones), it would not be clear why the embedded position could not be overtly filled.

Fourth, we already saw in (20), repeated here as (60), that if the accusative subject is pronounced inside the matrix clause, while the remainder of the embedded clause is leftwards scrambled, the result is ungrammatical.

This type of ill-formedness is usually attributed to a violation of the Proper Binding Condition (PBC): the trace that is left in the embedded clause after Dorj-iig ‘Dorj-ACC’ moves into the matrix clause can no longer be licensed after the remainder of the embedded clause itself moves to a higher position. This explanation only goes through if it is assumed that the embedded subject is base-generated inside the embedded clause and only later moved into the matrix clause.

The PBC is far from an uncontroversial issue, but it may be worth empirically comparing (60) with the linearly similar (61d). (61a) suggests that (61d) is not a hyperraising paradigm. In the latter, barring independent restrictions, nominative and accusative case alternate with each other. Nominative case is, in contrast, not available here (61a). Given the meaning of the matrix verb (persuade), it could be the case that the accusative in (61b) is a matrix argument. Consistent with this hypothesis is the possibility of filling the embedded subject position with an overt pronoun, as in (61c), which suggests that the empty category in the embedded subject position could be adequately analyzed as a dropped subject (pro). While this sentence is somewhat degraded, (61c) should be compared with (59), where we saw that a hyperraising accusative DP is not compatible with an embedded pronoun. Crucial for the purposes here, (61d), where the embedded clause is moved leftwards, is grammatical, even though the accusative DP is not moved along.

As observed by Bruening (2001; 2002) regarding similar data in Japanese, this can be explained if there is no trace in the leftwards scrambled clause in (61d).

Fifth, hyperraising in Mongolian cannot take place out of islands. (62c) contains examples of embedded coordinated clauses. The subjects of the coordinated clauses can be both nominative or both accusative. However, as (62c) shows, the accusative subject of one of the conjuncts cannot occupy a matrix position. This impossibility can be explained if hyperraising in Mongolian does involve raising and coordinated clauses are islands for movement.

Conditional clauses are also islands for hyperraising.¹⁶

The upshot is that accusative subjects in Mongolian that are pronounced in the matrix clause move (more precisely, hyperraise) there – the empty category ec in the data above can therefore be interpreted as a residue of movement.

Just like case assignment across a clausal boundary, hyperraising poses a challenge to assumptions about locality (i.e. the PIC in (30)). The edge-based analysis (42) can circumvent this problem too: the embedded subject moves to Spec-CP and from there it can not only be assigned accusative without violating the PIC, but also move into the matrix clause without violating this condition.

This analysis also helps explain why only accusative subjects can hyperraise. First of all, we must take note that local scrambling is a possibility in Mongolian.

Against this backdrop, consider what happens if we try to scramble a non-accusative subject across a finite CP. (66) shows that, if the subject of an embedded finite clause moves into the matrix clause while maintaining nominative case, the result is ungrammatical.

The same holds of objects, be they differentially marked (67) or not (68).¹⁷¹⁸

In (69), the accusative object is part of an embedded idiom, so that in (69b), the matrix accusative DP is clearly moved from the embedded clause. Similarly to what happens in (67) and (68), the result is also ungrammatical.

Finally, dative objects in the embedded clause cannot be pronounced in the matrix clause either.

One might ask at this point whether replacing or rewriting the case of an embedded argument after hyperraising or long distance-scrambling it would result in a grammatical sentence. In (71), an accusative DP placed in the matrix clause tries to be interpreted as the embedded goal argument, which would receive dative case in its base position (cf. (70)). The result is ungrammatical.¹⁹

What (66)–(70) show is that Mongolian does not allow for long distance scrambling, that is to say, the movement of an embedded argument into the matrix clause. But why is it possible for accusative subjects to move into the matrix clause, as we saw in examples like (52b)? Following Hiraiwa (2005) and Bondarenko (2017), the edge analysis in (42) can explain this possibility: moving the embedded subject to Spec-CP not only allows it to receive accusative case from the matrix v, but also allows it to locally move into the matrix clause, overriding a ban long distance scrambling.

Two clarifications are in order. First, a reviewer asks whether the A-movement of the accusative subject to Spec-CP and into the matrix clause is a separate operation or just an instance of local scrambling, illustrated on (64) and (65). The behavior of local scrambling suggests that hyperraising in Mongolian can be reduced to local scrambling. (73) shows that local scrambling feeds the creation of new antecedents for variable binding; (74) shows that local scrambling does not induce a WCO violation. As we are going to see below, these are some of the properties displayed by hyperraising.

Second, one may wonder at the nature of Spec-CP and the movement to that position. Spec-CP in the present analysis is not a designated escape hatch position; it is only created as a consequence of the valuation of the F-feature in COMP in (42). Assuming that case assignment is obligatory once the appropriate configuration is attained, the embedded subject that is moved to Spec-CP must be assigned accusative case, explaining why the retainment of nominative case is ruled out in (66). Furthermore, in keeping with locality conditions imposed on Agree, the closest DP that can satisfy the F-feature is the embedded subject. This blocks the movement of lower arguments like objects (67)/(68) and dative arguments (70).

In §2.3, we concluded that accusative subjects move to a higher position, analyzed here as Spec-CP. In the present section, we saw that the subject of an embedded finite clause in Mongolian can also hyperraise into the matrix clause. The edge analysis helped explain why this is a possibility. In §4, we examine the properties of the hyperraising landing site. We will see that this position dsplays some of the properties that are characteristic of A-movement, among them, the absence of weak crossover effects and of reconstruction for Condition C. If the analysis that hyperraising in Mongolian involves an intermediate Spec-CP position is correct, we would have to conclude that Spec-CP can be an A-position too. Before that though, I briefly comment on existing analyses of accusative subjects in Mongolian.

3.2 Comparison with previous analyses of accusative subjects in Mongolian

There are a few existing analyses of accusative subject DPs in embedded finite clauses in (some variety of) Mongolian. In this section, I briefly summarize and comment on the analyses by Guntsetseg (2010), Von Heusinger et al. (2011), and Bao et al. (2015).

Guntsetseg (2010) proposes that accusative subjects in Mongolian are instances of Differential Subject Marking (DSM). DSM is a type of differential case marking, which, in a rough definition, is the alternation between different types of morphology in a nominal or that cross-references a nominal, according to criteria like definiteness, animacy, specificity, etc. Guntsetseg argues that referentiality is one of the key components in regulating DSM in Mongolian and proposes that it has the particular function of signaling that the embedded subject is lower in referentiality than the matrix subject. Guntsetseg shows that speakers disprefer an accusative subject if it is higher in referentiality than the matrix subject, while showing no preference for an accusative form when the embedded subject is lower in referentiality.

A DSM analysis like Guntsetseg’s and the analysis here have different goals. The question that Guntsetseg aims at answering is, given the possibility of an embedded subject to surface with either nominative or accusative case, which option is chosen and what are the criteria to make the choice? Conversely, what I am trying to do here is to provide an explanation as to why these are options to begin with. Beyond this difference in perspective, a DSM analysis of this type would underdetermine some of the data examined here. We concluded in §2.3 that accusative subjects must occupy a higher position than their nominative counterparts. According to the DSM analysis by Guntsetseg, the occurrence of accusative case in the embedded subject is a consequence of a comparison with the semantic properties of the matrix subject, regardless of the position occupied by the subject.

In a similar vein, Von Heusinger et al. (2011) propose that the embedded subject of finite clauses is marked with accusative case because, Mongolian being a head-final language, the embedded subject may end up adjacent to the matrix subject. The embedded subject is marked with accusative case so that it can be dissimilated from the matrix subject. While this seems to be a reasonable way to view accusative DPs and the role of case in grammar, it is not readily compatible with some of the data inspected here, specially the binding data already mentioned above. Specifically, if accusative case only surfaced in the accusative subject in order for it to be distinguished from the matrix subject, the correlation between Condition A and Condition B effects discussed in §2.3 would not be accounted for straightforwardly.

Lastly, Bao et al. (2015) propose that accusative case in inner Mongolian is assigned by the embedded complementizer gej. A prediction that falls out from this kind of analysis is that the properties of the matrix clause should not matter to the availability of accusative case in the embedded clause, since the source of accusative case, the complementizer, belongs to the embedded clause itself. As we saw in the passive (28), accusative case in the embedded subject ceases to be a possibility in passives. The embedded complementizer though is unchanged in matrix passives, so that Bao et al. predict that the accusative variant in (28) should be grammatical, contrary to fact.

In the following section, we turn to the main goal of this paper, to inquire into the properties of a syntactic position like Spec-CP as afforded by the examination of hyperraising in Mongolian. We will see, among other things, that hyperraising can create new antecedents for binding and that it does not induce a weak crossover violation. These are properties that characterize A-movement. If the edge-based analysis in (91) is correct, the implication is that Spec-CP can be an A-position, contrary to the common assumption that it is inherently an Ā-position.

4.1 Hyperraising in Mongolian as a type of A-movement

Table 1 lists some commonplace properties that are used to distinguish between A- and Ā-movement. The table also anticipates the corresponding properties exhibited by hyperraising in Mongolian that will be examined in this section. As we can see, for the data available, this construction has the characteristic behavior of A-movement.

I should start with the properties regarding which I have no Mongolian data to offer. A-movement is well-known for creating new antecedents for anaphors. This property could not be surveyed here because the anaphors in Mongolian I am aware of seem to be subject-oriented. (76a) and (76b) show, by the means of a Condition C violation, that a dative goal can c-command into the theme argument. (76c) shows that, nevertheless, the goal argument cannot act as an antecedent to an anaphor contained in the theme. Interestingly, the fact that the anaphor is plural forces an interpretation where the subject denotes more than one person (despite the fact that this position is superficially occupied by a singular pronoun; this is informally represented with ‘+’). The goal argument, even though it is plural, is still not an appropriate antecedent.

If we try to embed an anaphor inside a matrix goal argument, the only valid antecedent is the matrix subject. Similarly to what we saw in (76c), the subject that is clause-mate with the anaphor in (77) must be plural.²¹

(78a) and (78b) provide analogous examples. (78b) shows additionally that being placed in the leftmost position still does not allow a hyperraised DP to bind a reflexive (here, the reflexive possessive -AA).

In sum, hyperraising does not create new antecedents for the binding of matrix anaphors, but there is likely to be an independent factor explaining this impossibility.²²

As for parasitic gap-licensing, I have no data to offer. As we know, there are several alternative analyses that must be ruled out before a PG-licensing-looking configuration can effectively be so characterized. Relevantly, preliminary data suggests that Mongolian allows for object-dropping. As such, I leave the intricacies of PG-licensing in Mongolian for future research.

On to the data available, a well-known condition that A-movement obeys is minimality, which states that this type of movement must target the closest nominal. A simple argument that hyperraising is a type of A-movement is thus supplied by the fact that only subjects can hyperraise. An object, for example, cannot, as we have already seen in (67), repeated below.

An indirect suggestion in favor of the claim that hyperraising in Mongolian patterns with other types of A-movement is provided by the compliance with the Activity Condition (Chomsky 2001). This condition prohibits nominals from undergoing A-movement if they have already been assigned case. As mentioned in fn. 6, hyperraising in general poses of problem for the Activity Condition, since it departs from the subject position of a finite clause, where it can arguably be assigned case. As such, it should be ineligible to undergo further instances of A-movement.

As also mentioned in fn. 6, there are several ways to avoid this problem. Among them, Kornfilt & Preminger’s (2015) analysis, could be appropriate for the Mongolian data examined here. Their analysis deals with Sakha, another Altaic language, like Mongolian. As a matter of fact, Kornfilt & Preminger state the problem not in terms of the Activity Condition, but, assuming Marantz’s (2000) dependent case framework, as a violation of the requirement that a nominal no longer count for the disjunctive hierarchy once it is assigned case. Based on Preminger’s (2014) formalization of a dependent case framework, Kornfilt & Preminger’s main proposal is that unmarked case is the exponent of an unvalued case feature. The nominative case that a hyperraising subject in e.g. Mongolian is assigned would be a type of unmarked case in Marantz’s hierarchy.

A prediction that falls out from this analysis is that subjects that are assigned some case other than unmarked case should be unable to undergo hyperraising. We have seen before possessive constructions in Mongolian (cf. (27b) and (32b)). In these constructions, the dative possessor induces a Condition C violation in an R-expression contained in the unmarked possessum:

If we embed a possessive construction like that in (80b) under the hyperraising verb khelekh ‘say’, the possessor cannot hyperraise, even though (80b) indicates that the dative possessor is higher than the unmarked possessum.

In order to apply Kornfilt & Preminger’s proposal to the data above, we could assume that dative case in a possessive construction is a type of lexical case. Accordingly, a possible explanation for (81b)’s ungrammaticality would be to say that a DP can hyperraise only if it is not marked for case. This is true of nominative subjects if nominative is type of no case at all, but not of dative case. (The unmarked possessum can hyperraise, as seen in (56).)

According to the assumptions made here, we could explain the correlation between case marking and the possibility of hyperraising in Mongolian as follows: it is possible to say that this type of construction obeys the particular version of the Activity Condition assumed here (Kornfilt & Preminger 2015), patterning with canonical instances of A-movement. This claim, of course, is only indirect, as mentioned in the outset, since it depends on very specific theoretical assumptions about case theory and the Activity Condition.²³

The absence of weak crossover effects yields a more compelling argument that hyperraising is a type of A-movement. In (82b), the embedded subject is a universal quantifier and it can be marked with nominative or accusative case. The matrix subject contains an internally complex possessive pronoun that is not subject-oriented (see Guntsetseg 2011 for details).²⁴ If the embedded subject surfaces below the matrix subject, as in (82a), the embedded subject universal quantifier cannot bind the pronoun in the matrix clause. However, if the accusative subject undergoes hyperraising and is pronounced in the leftmost position in the sentence, a variable binding reading obtains, (82b).²⁵

The position the hyperraised subject moves to must be an A-position, otherwise, the variable binding reading in (82b) would be undergenerated.

The same point is illustrated by the fact that the embedded subject can be coindexed with a pronoun attached to the matrix subject, regardless of whether the pronoun is a complex possessive (83b), a genitive pronoun (84b)/(85b), or a simple possessive (86b).²⁶

To summarize, hyperraising of the embedded subject to the leftmost position in the sentence does not disrupt the binding possibilities of the embedded subject that is below the matrix subject. In other words, hyperraising does not induce a weak crossover (WCO) violation, a defining property of A-movement. The overall conclusion then woult be that hyperraising in Mongolian is a type of A-movement.

Another property that hyperraising in Mongolian shares with canonical A-movement is the ability to fix an underlying Condition C violation. (87) shows that a matrix goal argument (tüün-d ‘3SG-DAT’) induces a Condition C violation on the embedded subject, irrespective of its case. (88) in turn shows that a Condition C violation can be repaired by hyperraising the embedded subject to the leftmost pronounced position in the sentence. The absence of reconstruction for Condition C is another characteristic property of A-movement.

Taken together, these data support the claim that hyperraising in Mongolian is a type of A-movement. If the edge analysis argued for here is correct, the implication is that Spec-CP can be an A-position too, challenging the commonly held assumption that it is intrinsically an Ā-position. This conclusion also allows us to identify what the F-feature assumed in (42) is. This is what we turn to in the section.

4.2 The F-feature as an A-type of feature

The data in the previous section help settle the question as to what type of feature the F proposed in (42) is. If F were a type of Ā-feature as in the analyses of subcategory (75b), the interpretive properties and/or grammaticality of the data above could not be accounted for. The reason is that Ā-features in a complementizer would engender a Spec-CP of the canonical Ā-type. Hyperraising in Mongolian was proposed here to pass through Spec-CP and, additionally, it was just shown to feed phenomena of the A-type. If we applied analyses like those in (75b) to Mongolian, the prediction is that the data above should be instances of improper movement (schematized in (2) above) and therefore ungrammatical, contrary to fact.

The A-movement-based edge analyses in (75a) fare better: according to these analyses, Spec-CP is a type of A-position. Assuming that the edge analysis pursued here is correct, hyperraising through Spec-CP in Mongolian is correctly predicted to not incur any violation of the ban on improper movement.

But what does it mean to say that Spec-CP can be an A-position? Implied or explicitly stated in the analyses in (75a) (Tanaka 2002; Takeuchi 2010; Obata & Epstein 2011; Wurmbrand 2017; To appear) is that syntactic positions are defined in terms of the features that create them (Obata & Epstein 2011; Van Urk 2015). Spec-CP can be an A-position if the features in the complementizer that create it are a type of A-feature. Takeuchi and Wurmbrand specifically propose that the features in the complementizer are φ-features. Takeuchi, based on Chomsky (2008), assumes that a complementizer can bear φ-features as a consequence of the failure of T to inherit these features from the complementizer (cf. Legate’s 2011 under-inheritance). Wurmbrand follows Van Urk (2015) and proposes that φ-features are a possible specification of a complementizer.

Van Urk argues at length against the assumption that syntactic positions in general should be classified as being inherently A or Ā. Rather, Van Urk proposes that they should be defined by the features that create them. For instance, Spec-CP is an Ā-position not because this is its intrinsic property, but because filling it was triggered by e.g. a Wh-feature. Similarly, Spec-TP does not have to be an A-position inherently either; it is defined as such because it is created by φ-features in T. Van Urk’s main thesis is stated as follows:

Van Urk’s proposal finds precedents in Bruening (2001), Chomsky (2007), and Obata & Epstein (2011). Bruening proposes that a variant of a hyperraising construction in Japanese and in Passamaquoddy that is similar to the Mongolian data examined here can be analyzed by base-generating a DP in an embedded Spec-CP and then moving it to the matrix clause. Bruening shows that this type of movement also feeds the creation of new antecedents for binding, leading us to assume that Spec-CP can be an A-position.

Obata & Epstein, following Chomsky (2007), also claim that syntactic positions should be defined by the features that create them. Specifically, Obata & Epstein define A- and Ā-positions as follows:

(90) is particularly useful for the present analysis because it allows us to further specify what the F-feature posited here is. Assuming Obata & Epstein’s formalization in (90), I propose that the subject of embedded finite clauses in Mongolian moves to Spec-CP in order to value φ-features in the complementizer gej, as represented in (91).²⁷

As hinted at above, the proposal that hyperraising can be triggered by φ-features in the complementizer is by no means novel, Bruening (2001; 2002), Tanaka (2002), Obata & Epstein (2011), and Wurmbrand (2017; To appear) having spearheaded this type of approach (see also Zyman 2017; 2018). Outside of hyperraising, this proposal has been put forward too. For instance, Shlonsky (1992; 2002) observes that a resumptive pronoun (boldfaced below) in Palestinian Arabic constituent questions is obligatory if the gap is in the object position (92b), but forbidden (93b) if it is the highest subject position (see more details in Shlonsky 1992; 2002). In the latter, a gap (represented with e) is required.

In order to account for this contrast, Shlonsky’s (1992; 2002) main proposal is that resumptive pronouns only occur as last resort strategies when Wh-movement is impeded. To motivate the impossibility of Wh-movement in (92b), Shlonsky proposes further that the complementizer ʔilli is lexically specified in such a way that its Spec position is defined as an A-position, as opposed to an Ā-position. Movement from the subject position (Spec-TP) to Spec-CP is expected under these assumptions, which is why a last resort resumptive pronoun is not called for in (93b). However, movement of the object to the same landing site incurs a minimality violation, hence why a resumptive pronoun is required in (92b). Ultimately, Shlonsky proposes that Spec-CP in Palestinian Arabic is an A-position because it is created by φ-features in ʔilli.

Going back to Mongolian, because the creation of Spec-CP in the variety of hyperraising found in this language (see §4.1) is proposed to be triggered by φ-features, this is defined as an A-position too. If the subject that moves there subsequently hyperraises into the matrix clause, this target position being one that has properties that characterize A-positions, the ban on improper movement is not violated. A sentence like (84b) above can be represented as in (94), where A-movement to the final landing site is preceded by movement to Spec-CP, a type of A-movement itself (cf. improper movement in (2); for ease of comparison, the tree in (94) is head-initial).

(94) schematizes the main claim I make in this paper, namely, the claim that movement through Spec-CP can feed A-movement without violating the ban on improper movement. Alternatively said, (94) illustrates the proposal that movement through Spec-CP can be an instance of proper movement, that is, a type of A-movement that can feed a subsequent instance of A-movement. A similar conclusion has been reached previously by Zyman’s (2018) analysis of hyperraising to object in Janitzio P’urhepecha.

In the next section, I will draw a comparison between hyperraising and covert Wh-movement in Mongolian, in order to highlight that the former is a type of A-movement through Spec-CP, while the latter is a canonical example of Ā-movement.

4.3 A vs. Ā: A comparison between hyperraising and Wh-movement

In the preceding section, hyperraising in Mongolian was shown to display several properties that are attributed to canonical instances of A-movement, like the creation of new antecedents for binding and the absence of reconstruction for Condition C. In this section, we examine Wh-movement, which displays the opposite behavior. First, though, it is necessary to show that Wh-phrases do indeed involve covert movement, this being a Wh-in situ language.²⁸