Successive Cyclicity in DPs: Evidence from Mongolian Nominalized Clauses

A well-established property of long-distance movement is that it is successive-cyclic: phrasal movement of an XP from its base position to the one where it is pronounced takes place in a series of shorter steps. This punctuated nature of movement is often tied to phasehood. Phases are taken to be special in that they (a) may provide intermediate positions along the path of successive-cyclic movement where moving XPs can stop off and (b) force XPs to land in these positions by virtue of being opaque domains otherwise. By now, there is a growing body of evidence that long-distance movement stops off at the edge of each intervening CP (Henry 1995, McCloskey 2002, Torrego 1984) and vP (Bruening 2001, Rackowski and Richards 2005, Van Urk 2015) (see Citko 2014 for an overview). However, it is debated whether DP, another purported phasal domain, hosts escape hatches and allows intermediate movement through its edge (Bach and Horn 1976, Chomsky 1973, Cinque 1980, Gavruseva 2000, Giorgi and Longobardi 1991, Matushansky 2005, Svenonius 2004, Szabolcsi 1994, Tellier 1991). On the one hand, Complex Noun Phrase Constraint effects as in (1) may be taken to show that DPs lack an escape hatch.

On the other hand, left-branch extraction facts crosslinguistically have been argued to show the opposite (Cinque 1980, Gavruseva 2000, Giorgi and Longobardi 1991, Szabolcsi 1984, 1994). For instance, Szabolcsi (1994) shows for Hungarian that only possessors that appear to the left of determiners when nonextracted can undergo movement from their containing DP. She and others (see, e.g., Gavruseva 2000) have tied this to the availability of an escape hatch position within DP. Similarly, in Romance, only those arguments that can be possessivized can move out of the DP, which has also been taken to show that XPs must pass through Spec,DP in order to move out of the phase (Cinque 1980, Giorgi and Longobardi 1991). However, such an interpretation of these facts is not uncontroversial. Left-branch extraction data have been analyzed by others as showing instead the absence of a D layer altogether (Bošković 2005, 2014, Uriagereka 1998). Thus, the question remains: does DP allow—and require—movement of phase-internal material through its edge?

In this squib, I present novel evidence from Mongolian (Mongolic) showing that movement happens successive-cyclically from at least some DPs. Nominalizations in Mongolian require Ā-movement out of them to stop off at Spec,DP. Supporting evidence comes from unexpected interactions between embedded subject case and movement. Subjects of these nominalized clauses can ordinarily receive nominative (NOM), genitive (GEN), or accusative (ACC) case. The subject of a nominalized clause receives ACC case if it occupies Spec,DP, but movement from within the clause disrupts this case possibility. Specifically, (a) ACC on the subject of a nominalized clause is impossible when a nonsubject undergoes movement out of that DP, and (b) ACC is impossible on the subject of any intermediate nominalized clause that is crossed by movement. I argue that the (un)availability of ACC case marking is a reflex of successive cyclicity in DPs: when ACC on a nominalized clause subject is blocked, it signals that the edge of that DP has been targeted for intermediate movement. More broadly, these results lead to the conclusion that Spec,DP serves as a landing site for intermediate movement in at least some DPs and suggest that phases of any category can, in principle, provide escape hatches for movement out of them.

Subordinate clauses in Mongolian are ordinarily headed by the complementizer gež (2). But in addition, a range of embedding predicates, including verbs of perception (xarsax ‘see’, sonsax ‘hear’), (certain) cognitive factives (medex ‘know’, olž medex ‘find out’), and verbs of saying (xelsex ‘tell’), take nominalized complements. Though an overt nominalizer is absent, the nominal nature of these clauses is illustrated by the fact that they are obligatorily case-marked (3), can be complements of prepositions (4), and can appear in subject positions (5). CPs in this language do not show any of these properties.

Subjects of nominalized embedded clauses may be case-marked in one of three ways: with NOM (unmarked), GEN (-iin/nii), or ACC (-(ii)g).

Though these forms are seemingly in free variation, when we look beyond the basic cases, we find syntactic environments where one or the other case morphology is blocked. I take these asymmetries to reflect differences in the syntactic position of the nominal in question, as schematized in (7).¹ Evidence for such structural differences among differently case-marked subjects comes from specificity effects, NPI (negative polarity item)-licensing facts, and variability in case possibilities depending on the syntactic position of the clause. I elaborate on these arguments below.

Case on subject nominals varies as a function of specificity. For instance, a nonspecific indefinite subject of a nominalized clause can only bear NOM (8). Pronouns, on the other hand, show the opposite pattern in disallowing NOM (9).²

These patterns provide indirect evidence that GEN/ACC-marked subjects are structurally higher than NOM subjects. More specifically, a NOM subject remains in the vP-internal position where it is generated, whereas GEN/ACC subjects move out of this position (see Asarina 2011 and Gribanova 2017 for similar claims about Uyghur and Uzbek, respectively). Following proposals by Diesing (1992), Diesing and Jelinek (1995), and much work since, the interpretational differences between specific and nonspecific nominals can be treated as a consequence of their relative position with respect to an Existential Closure operator. The specificity effects discussed above follow if Existential Closure occurs immediately above vP (and below nP), such that vP-internal subjects remain in the scope of this operator. Any indefinite that receives an existential interpretation must be in the scope of Existential Closure. Moreover, since Existential Closure is taken to be unselective (Heim 1982), any nominal that introduces a free variable and must not receive an existential interpretation (e.g., referring pronouns) must move out of its scope.

It is important to note that while specificity and case marking are often linked, case-marking differences cannot be reduced entirely to the semantic property of specificity (contra, e.g., Guntsetseg 2010, 2016). There are certain nonspecific indefinites—namely, those formed with the focus particle -č—that can be nonnominative.³

In the presence of clausemate negation, these indefinites receive an NPI interpretation. Crucially, in such environments, the subject can only bear NOM.

This restriction follows from the posited positional asymmetry between NOM and GEN/ACC subjects. The negative morpheme -gui, when it appears in nominalizations, is structurally higher than vP but, crucially, lower than nP. This is a direct consequence of the selectional properties of the morpheme, which is restricted to the verbal/clausal domain and cannot compose with nominals in the first place. As a result, only a vP-internal indefinite—which is NOM-marked—would be in the right position to display NPI-hood.

GEN and ACC subjects pattern together in most environments, but, crucially, the availability of ACC on a subject is limited to nominalizations in complement position. When the nominalized embedded clause appears in subject position, ACC is unavailable.

This suggests that the licensing of ACC on embedded subjects depends on the presence of some higher-clause element. An ACC-marked subject has not, however, overtly raised into the higher clause, as argued at length in Guntsetseg 2016, von Heusinger, Klein, and Guntsetseg 2011, and Klein, Guntsetseg, and von Heusinger 2012. For instance, subjects of adjunct nominalizations can bear ACC.

Moreover, an ACC-marked subject cannot be separated from the rest of the nominalized clause by matrix elements (14) or stranded when the nominalized clause containing it undergoes scrambling (15). Thus, what we have here is not classical Raising-to-Object.

The resulting picture, then, is one where ACC-marked subjects must stay within the clause, but also establish a dependency with some higher-clause element that licenses ACC case. Given the phasehood of DPs, the only position where material internal to them is visible for operations outside the phase is the edge, that is, Spec,DP. Therefore, I propose that ACC case can be licensed only on nominals that move to this peripheral position. Thus, whereas both GEN- and ACC-marked DPs have vacated their vP-internal position, ACC subjects, by virtue of being at the DP edge, are structurally higher than GEN subjects.

For the purposes of this squib, it is not important how subject case is licensed, as long as the DP in question is forced to be at the phase edge to receive ACC. Two prominent families of case theories—head-licensing approaches (Chomsky 1981, 2000, 2001) and configurational approaches (Baker and Vinokurova 2010, Bittner and Hale 1996, Levin and Preminger 2015, Marantz 1991)—make the same predictions regarding this requirement. In theories that attribute case assignment to dedicated functional heads, ACC case on embedded subjects may be viewed as being exceptionally licensed by a higher-clause v. On a configurational approach, ACC case is a manifestation of dependent case, assigned to a DP if it is c-commanded by another DP within some local domain. ACC on embedded subjects could arise because the nominal in question may be in the same case competition domain as the higher-clause subject (Baker 2015, Baker and Vinokurova 2010). Importantly, on either approach, subjects receiving ACC must be local to some higher, clause-external element, either the higher v or the higher subject, a requirement that is met only by subjects at the phase edge.⁵

To sum up, I have suggested that NOM-marked subjects of nominalized clauses remain within the vP domain, whereas GEN- and ACC- marked subjects occupy higher, derived positions. Because ACC licensing requires that the relevant nominal be accessible to elements outside of the DP phase, I have proposed that ACC subjects raise to Spec,DP.

In this section, I will argue that the availability of ACC case is a diagnostic of successive cyclicity in the nominalized clauses examined here. The logic of the argumentation is as follows. Suppose a nominalized clause that otherwise permits ACC subjects loses its ability to host such DPs when movement has taken place. This would indicate that the moving XP lands in Spec,DP on its way to its final landing site, in turn making this ACC position unavailable for subjects.

My evidence relies on contrastive topicalization involving the marker bol. (16) shows that bol-topicalization reconstructs for binding. Anaphors in Mongolian must be c-commanded by a local antecedent, as shown by the ungrammaticality of (16a); topicalization does not disrupt otherwise grammatical binding relations, as shown in (16b) vs. (16c).

The examples that follow illustrate the Ā-properties of topicalization—in particular, its island sensitivity and its ability to take place long-distance. Relative clauses in Mongolian are introduced by a null relativizer, but word order frequently makes clear that relativization has taken place (17). Relative clauses are islands in this language, and as shown in (18a), topicalization cannot take place out of them. Topicalization that violates the Coordinate Structure Constraint is also banned (18b). Finally, as shown in (18c), topicalization can take place across a clause boundary.

Constituents inside nominalized embedded clauses can be extracted by topicalization, but this movement interacts with subject case, whether or not it is the subject itself that is topicalized. First, if an embedded subject of a nominalized clause is topicalized, it must bear ACC case. The baseline in (19) shows that the subject of the nominalized clause can be marked NOM, GEN, or ACC.⁶ In (20), however, the embedded subject undergoes topicalization and only ACC is possible.⁷

Second, if a nonsubject is topicalized, the embedded subject cannot bear ACC; compare the ill-formed (a) variants in (21) and (22), where the subject of the nominalization bears ACC, with the (b) and (c) variants, where the subject is unmarked and GEN-marked, respectively.

Finally, and strikingly, extraction of a constituent from an embedded nominalized clause has ramifications for the case on the subject of every intervening nominalized clause along the path of movement. Specifically, the subjects of these intervening clauses cannot bear ACC, even though movement did not originate from within them. The sentences in (23), for example, involve the topicalization of a deeply embedded object nominal, but the variant where Bat is marked ACC is unacceptable. Crucially, ACC on intermediate subjects is not blocked in (23b), which is identical to (23a) except that no movement has taken place. This asymmetry confirms that it is indeed topicalization from the lower clause that blocks ACC case of the intermediate nominalized clause.⁹

To capture this observed correlation between long-distance topicalization and subject case, I propose that in Mongolian nominalized clauses, there is a single edge position, Spec,DP, where any nominal that needs to be visible to phase-external operations must move. If there is a single edge position accessible to phase-external processes, and if this is also the only position in which the subject of a nominalized clause receives ACC case, the correlations discussed above are predicted and explained.¹¹

A number of authors have argued that the binding domain for Principle A should be stated in terms of phases (Hicks 2009, Lee-Schoenfeld 2008, Safir 2014): an anaphor can be bound outside of its own phase only if it is located at the edge of that phase. If what I have argued thus far is on the right track, we expect that only XPs that occupy Spec,DP—that is, ACC-marked subjects—are able to be bound by an element outside of the nominalization. Moreover, if movement out of a nominalization must proceed through Spec,DP, as argued in section 3.1, it is also expected that such cross-phasal binding can never cooccur with topicalization. Below, I show that these predictions are borne out.¹²

To express possession on a DP whose possessor is an element within the same sentence, Mongolian utilizes the reflexive possessive suffix -AA (subject to vowel harmony).

Cross-clausal binding of a reflexive possessive-marked nominal inside a nominalized clause is possible, but in such cases, NOM is not possible (25). Moreover, if the antecedent is the subject of a higher nominalization, it must be NOM-marked. Thus, in (26) the subject of the intermediate embedded clause, which binds the reflexive possessor-marked subject of the most deeply embedded nominalization, cannot appear with GEN or ACC case.

What these patterns show is that reflexive-possessive binding in Mongolian requires a highly local relation between the antecedent and the reflexive. In (26), for example, this relation can only be successful if the antecedent remains within the higher vP phase and if the anaphor is at the edge of the lower phase, Spec,DP.

Crucially, topicalization out of a nominalized clause makes binding of a possessed subject impossible. As we see in (27), irrespective of the case morphology on the subject, reflexive possessor-marking renders the construction ungrammatical. The way to rescue this structure is to use the nonanaphoric third person pro-form -n’, as shown in (28).

These data thus corroborate the present account. Movement out of nominalizations always proceeds through Spec,DP, disrupting any other process requiring the use of the edge.

In this squib, I presented novel evidence from Mongolian nominalizations that extraction out of DPs involves intermediate movement through every phase edge on the way to the landing site, making DPs no different from CPs or vPs with regard to successive cyclicity. More generally, these data point to a perspective on phases where there are no fundamental asymmetries across phasal categories with regard to the availability of a landing site for intermediate movement. Of course, even in Mongolian, extraction from other kinds of DPs tends to be harder than from CPs and vPs, a fact that calls for explanation. But it is beyond the scope of this squib to delimit the conditions under which extraction is possible. What I hope to have shown is that when movement out of a DP is possible, it happens in the same way as movement out of CPs and vPs—in a series of short steps. Having established that, the next step is to identify general architectural or language-specific properties that make movement out of DPs much more severely constrained than movement out of these other phases.

All uncredited data are from original fieldwork with two informants, Urandari Byambadalai and Buyandelger Manduhai; I am grateful to them for sharing their language with me. In addition, Urandari Byambadalai reviewed the transliterations for accuracy and conformity to orthographic conventions. Thank you to three reviewers for detailed comments on versions of the squib. My thanks also to David Pesetsky, Norvin Richards, and Coppe van Urk for helpful comments and feedback.