2.1 Immobility

DP/NP approaches operate under the assumption that arguments are free to enter the derivation as DPs or NPs, while the latter will only be licensed in a PNI scenario, that is in a context where the verb and the object denote a conventionalized or at least frequently occurring event. The size of the noun phrase correlates with meaning, mobility, and case. Case marking is often tied to a [D]-feature (Dayal 2011; Barrie & Li 2015), while the lack of a DP shell reduces arguments to semantic objects of type ⟨e,t⟩. This creates the need either for incorporation verb denotations (van Geenhoven 1998; Dayal 2011) or a new compositional mode (Chung & Ladusaw 2004) to combine properties and verb denotations.

Important for the present study is the connection between a reduced nominal domain and apparent immobility. Frey (2015: 243) for example argues that PNI-ed objects in German constitute NPs and denote properties, thereby requiring complex predicate formation with PNI verbs to ensure semantic composition, which in turn leads to a general compactness requirement.³ A very different account is provided by Massam (2001) for Niuean, an Oceanic language with obligatory verb-initial word order. Consider the minimal pair in (5), where (5b) constitutes the PNI scenario. Both verb and object have to be adjacent, while the object is stripped off its number and case marking.

Massam (2001) proposes that pseudo-incorporated arguments are NPs and therefore do not require case. Hence, they do not move into a case position, which is projected in a dedicated functional layer outside of the verbal domain. The analysis is demonstrated in (6b), based on the PNI context in (6a). Since the object does not move for case, the subject moves to the specifier of AbsP, where it is assigned absolutive case. VP-fronting is triggered by the feature [PRED] on I. PNI-ed objects are obligatorily interpreted as non-specific indefinites, which Massam takes to be a consequence of the NP status.

López (2012) attributes movement restrictions for Spanish low-scope indefinites and bare plurals to the assumption that they form syntactic phrases which do not constitute phases. He analyzes accusative case as the spell-out of a K head which, if present, introduce a choice function (Reinhart 1997) that enables flexible scope. K heads project KPs which constitute phases and thus are able to undergo scrambling. Any nominal argument smaller than a KP, i.e. DPs, NumPs, and NPs denote properties, must remain in their base positions and can only be interpreted via the compositional rule Restrict (Chung & Ladusaw 2004), a semantic operation tailor-made for PNI contexts which combines properties with verb denotations that apply to individual type arguments.

2.2 Adjacency

While head movement accounts were predominantly proposed for noun incorporation (Sadock 1980; Baker 1988; 1995; 2009; Chung & Ladusaw 2004; Baker et al. 2005),⁴ Baker (2014) extends this analysis to pseudo-incorporation, taking Sakha and Tamil as case studies. According to Baker, pseudo-incorporation is different from noun incorporation in that it has to be string vacuous, i.e. noun and verb have to be linearly adjacent. Crucial for Baker’s analysis is that head movement is not enforced by feature checking, thereby making its application completely optional. PNI semantics, however, can only be generated if N and V form a complex head, that is N denotes a property only if it is dominated by a V node (Baker 2014: 20–21). Baker adopts the copy theory of movement and follows Nunes (2004) in that lower copies are deleted due to the presence of uninterpretable features. Since the type of head movement shown in (7b) is not feature driven, either one of the copies can be spelled-out, leading to potential linearization issues if movement is not string vacuous. Thus, Baker derives surface adjacency between N and V through the avoidance of ordering paradoxes at PF. In order to explain the case loss, Baker follows Baker et al. (2005) by making use of a parameterized deletion rule which removes Φ-features on traces of head movement. Along with Φ-features, Baker assumes case information can also be lost.

Evidence for his proposal comes from the ban on scrambling of caseless objects in Tamil and Sakha, which is in line with our observations for Tamil in section 4.1. Baker, furthermore, claims that low manner adverbs cannot intervene between the verb and the object, shown in (8).

Resultative structures provide further evidence for surface adjacency. He uses the word order contrast between (9a) and (9b) to illustrate that surface adjacency can be achieved within the course of a derivation by moving the resultative PP out of the intervening position. Baker suggests a similar analysis for (9c), although in case of adverbial resultatives an additional movement step as in (9b) is disallowed for independent reasons.

The second type of adjacency approach I want to take a closer look at is the head-to-head requirement between verbs and PNI-ed arguments proposed by Levin (2014; 2015) to account for PNI in Tamil, among other languages. The main empirical observation Levin wants to capture is that the highest nominal projection within these arguments seems to require surface adjacency with the lexical verb. Levin redefines the case filter as a restriction on size rather than a restriction on feature valuation. All categories must be part of a complete extended projection which in case of noun phrases is a KP.

(11)	Levin’s structure of the noun phrase:

If noun phrases are merged as anything less than a KP, say a DP or an NP, the head of the highest nominal projection must be licensed by forming a complex head with the lexical verb via adjunction as part of post-syntactic local dislocation (Embick & Noyer 2001). The analysis is schematically shown in (12) for Tamil, where • encodes immediate precedence. Since V and N are linearly adjacent, N adjoins to V at PF.

Crucially, local dislocation obviates the case filter since the nominal becomes part of the verbal projection. Since it is the K head that carries the case feature, non-KPs are not case licensed, enforcing non-KPs to stay linearily adjacent to V, as this is the only configuration where local dislocation is permitted. The ban on displacement as well as the intervention of adverbs (8) and resultative PPs (9) follow directly from the linear adjacency requirement.

To sum up, both adjacency and immobility accounts predict that PNI-ed arguments and verbs behave like a unit, preventing the PNI-ed object to dislocate from its base position. As the next section will show, both kinds of approaches turn out to be empirically inadequate, once a larger data set is taken into account.

3.1 No surface adjacency in Tamil

As was presented in section 2.2, caseless objects in Tamil are argued by Baker (2014) and Levin (2015) to only be licensed in positions linearly adjacent to the verb. Baker (2014) uses low manner adverbs and resultative constructions to illustrate the compactness constraint. Counter-evidence against surface adjacency comes from focus adverbs/particles which occur to the right of the nominals they scope over, and thus potentially in between PNI-ed objects and verbs. As it turns out, PNI scenarios do not block the occurrence of such focus operators. Both maṭṭum in (13) and kuuţa in (14) intervene between a caseless object and the verb, contrary to what Baker (2014) and Levin (2015) predict. Even if the focus adverbs were to be analyzed as part of the nominal domain, pustagam would not be able to undergo string vacuous head movement to V in Baker’s sense. Likewise, pustagam constitutes the highest nominal projection but is not adjacent to V, thereby violating Levin’s head-to-head adjacency constraint.

Moreover, the study was not able to verify Baker’s surface intervention effects with low manner adverbs, recall (8). All four Tamil speakers agreed on the acceptability of veegamaa placed between a caseless bare noun and a verb.⁵ The speakers do, however, agree with the judgements of the resultative structures discussed in section 2.2. In light of the data above, we cannot attribute the unacceptability of such examples to violations of surface adjacency. One alternative explanation can be given in terms of different base orders. Under the assumption that locative PPs in (9a) and (9b) are introduced by a high applicative head (Marantz 1993), the surface order of such structures would have to come about by scrambling the PNI-ed object across the PP – an illicit movement step according to Baker (2014) as well as the current approach. The underlying structure for (9a) is sketched in (15), where the movement step is assumed to target an inner specifier of vP. Consequently, resultatives like (9b) in which the PP precedes the direct object are acceptable since they constitute the base order where the PNI-ed object still occupies its base position.

The unacceptability of the adverbial resultative in (9c) is quite likely due to the fact that resultatives necessarily include a control structure, as it is often discussed for adverbial small clauses co-occuring with transitive predicates (von Stechow 1995; Beck & Johnson 2004). Since the subject of the small clause is also an object of the matrix predicate, syntactic requirements enforce a control relation between the PNI-ed object and silent PRO, shown in (16).

Crucially, PNI-ed objects in Tamil are banned from acting as controllers generally, as (18) demonstrates. Whatever prevents PNI-ed objects from acting as controllers will also serve as an explanation for illicit PNI in adverbial resultatives.⁶

Summing up, the incompatibility of resultative structures with PNI-ed objects in Tamil can be accounted for on independent grounds without the need for a surface adjacency requirement. Together with the insight that focus adverbs as well as manner adverbs can intervene between caseless bare objects and verbs, PNI compactness based on surface adjacency becomes highly unlikely.

While PNI-ed objects do not obey a surface restriction, they do in fact seem to be severely restricted in their movement capacity, as the examples in (19) show. Only the base position in (19a) is acceptable, whereas short scrambling and movement across the subject is not licensed. Neither is long scrambling an option, as (19b) shows.

To sum up, PNI-ed objects in Tamil are not able to undergo scrambling, yet they do not have to be surface adjacent to the verb.⁷

3.2 Scrambling in Turkish

While Turkish displays short, intermediate, and long scrambling properties (Kornfilt 1997; Termücü 2005; Öztürk 2005; Işsever 2007; İşsever 2008; Akan 2009; Jiménez-Fernández & İşsever 2012), caseless bare nouns are traditionally assumed to be restricted to occur in their base positions (Erguvanlı 1984; Dede 1986; Enç 1991; Kornfilt 2003; Aygen 2007). Recent studies, however, provide data which question this generalization. Kornfilt (2003: 152) mentions in a footnote that caseless bare nouns can be non-adjacent to V in colloquial speech. Moreover, İşsever (2003) shows that caseless nouns can extrapose, while Öztürk (2009) and Gračanin-Yüksek & İşsever (2011) provide examples of clause-internal scrambling, for example as is shown in (20) for intermediate scrambling. In (21), we see that caseless bare nouns are acceptable preceding and following the indirect object.

The availability of long scrambling is controversially discussed in the literature. Gračanin-Yüksek & İşsever (2011) claim that long scrambling is blocked for caseless bare nouns, based on the data in (22). Jo & Palaz (2018), on the other hand, provide data where long scrambling is clearly acceptable, see (23). Note that the authors make use of different matrix verbs. The speakers of this study agree with the judgements in (23) and (22a), but they disagree on the acceptability of (22b).

Turkish long scrambling seems to be sensitive to whether the matrix verb qualifies as a bridging verb. Verbs of saying often allow long distance extraction more readily than e.g. low frequency factives like regret, while some languages also block highly frequent factives like know (Kluender 1992; Hawkins 1999). The acceptability of long scrambling with a typical bridging verb like söye ‘say’ in (24) confirms this hypothesis. Cross-linguistic observations concerning long distance extraction ascribe söye and düşün a better chance for enabling long scrambling since they qualify as prototypical bridging verbs. Hence, it can be concluded that the unacceptability of the structures in (22) is not tied to pseudo-incorporation.

In contrast to Tamil, Turkish bare objects are entirely free to scramble short, intermediate, and out of a finite clause, whether they are case-marked or not. This cross-linguistic diversity does not fall out from DP/NP accounts which adhere movement restrictions to the reduced NP status. The next section will introduce yet a different movement pattern for PNI-ed objects, found in German.

3.3 Topicalization in German

So far, we have limited our investigation to PNI languages showing a correlation between case loss and scope inertness. German has been argued by Frey (2015) to exhibit PNI effects for bare plurals and non-specific indefinites, albeit without an effect in case marking. Frey’s diagnostics consist of scope inertness of certain noun types in combination with certain positional restrictions. He observes that there is only a small class of arguments, made up of non-specific indefinites and bare plurals, that can follow manner adverbs and negation, shown for the adverb wunderbar in (25) and (26). The indefinite in (25a) occurs next to the verb and can only be non-specific, proven by the infelicity of adding bestimmt ‘certain’ in (25b). If the indefinite scrambles out of the verb phrase, it receives a specific reading, see (26a), where the underlying structure is given in (26b).

Moreover, non-specific indefinite objects are unable to scope above a universally quantified subject if they follow a low manner adverb, shown in (27).

The contrast in (28) provides a similar observation for bare plurals, see (28b) with the sketched structure in (28c). Like non-specific indefinites, bare plurals are licensed in V-adjacent position (28a) but not in a derived position (28b).

The underlying structures in (26b) and (28c) suggest that non-finite verbs on their own can only occur sentence initially as part of remnant VPs, where objects vacate VPs prior to VP-topicalization. This first step is the cause for the unacceptability of (28b) and the specific reading of the object in (26a), as PNI-ed objects are required to undergo scrambling so that they can be left stranded by the VP.

Frey (2015) provides a different analysis. He proposes that bare plurals and non-specific indefinites can form a complex predicate with the verb in PNI contexts. The unacceptability of (28b) and the specific reading of the object in (26a) under his account are simply violations of the PNI compactness constraint since they would require excorporation of the verb. Unpredicted by his analysis, however, is the fact that PNI-ed objects can topicalize, shown in (29a) with the structure given in (29b).

The contrast between (29a) and (28b) forces Frey to an ad hoc stipulation. While V cannot excorporate out of a complex PNI predicate, objects are able to do so. The current proposal, on the other hand, takes the contrast between (29a) and (28b) to be indicative of different types of movement operations PNI-ed objects can undergo. The discussion in this section suggests that PNI-ed objects are prevented from scrambling but are free to undergo topicalization. In this sense, movement of PNI-ed objects in German is not as restricted as the movement pattern in Tamil, yet also not as unrestricted as the Turkish pattern.

3.4 Problems with cross-linguistic variation

A cross-linguistic comparison between Tamil, Turkish, and German provides evidence against the compactness constraint argued for in section 2. The observations create an insurmountable problem for DP/NP accounts which explicitly argue for the NP status based on observations of complete immobility and/or surface adjacency of PNI-ed objects with the verb (Massam 2001; Dobrovie-Sorin et al. 2006; López 2012; Levin 2015; Barrie & Li 2015). Surface adjacency is also predicted by head-movement accounts to PNI (Baker 2014; Kornfilt 2003)⁸ as well as complex predicate formation (Aydemir 2004; Frey 2015). In contrast, DP/NP accounts that do not make reference to movement restrictions of PNI-ed objects in the first place (Öztürk 2005; 2009; Dayal 2011) cannot explain why PNI generally leads to mobility restrictions, in comparison to non-PNI scenarios. The current account is able to overcome these issues by drawing a parallel to VP-movement patterns which are commonly known to vary from language to language.

4.1 Tamil & Mongolian

Section 3.1 established that PNI-ed objects do not undergo short, intermediate, or long scrambling in Tamil. PNI-ed objects in Mongolian exhibit identical movement restrictions. Whereas case-marked indefinite objects scramble freely (Janhunen 2012; Guntsetseg 2016; Fong 2019),⁹ caseless indefinite objects cannot scramble across an indirect object (31b), nor are they able to precede the subject (32b) or undergo long scrambling (33b).

The task of this section will now be to demonstrate that neither Tamil nor Mongolian allow for VP-scrambling. In order to investigate the possibility of VP-movement in Mongolian, we test for the acceptability of postverbal constituents. The Mongolian speakers of this study rejected any type of postverbal constituent, ranging from adverbials over indirect objects to subjects, shown in (34).

As the reader might have noticed, we are faced with the problem of isolating the VP-constituent from the affixed tense morphology in the scrambling data above. Thus, a likely explanation for the fact that we do not find evidence for VP-movement might be that obligatory V-to-T movement bleeds any dislocation of overtly headed VPs from their base positions. There is, however, positive evidence for an alternative analysis that requires post-verbal constituents to have a specific information structural profile, which is not present in (34). Guntsetseg (2016: 25) reports that postverbal elements are generally disallowed, except when they are separated by a pause and denote “additional information”, see also Poppe (1951: 112) and Binnick (1979: 122). Consequently, the pragmatic function of these structures have been described as an afterthought by Janhunen (2012: 228) and Öztürk (2013: 192). The following contrast in (35) highlights the information structural aspect of the construction where | signals a pause.

Öztürk (2013) offers a bi-clausal analysis in which postverbal constituents in Mongolian are part of a separate clause, adjoined to the preceding one. The second clause is partially deleted under phonological identity with the antecedent clause. Within the first clause, the counterpart of the post-verbal constituent is realized as pro. Similar analyses have been carried out for Japanese (Tanaka 2001), as well as Dutch and German (Ott & de Vries 2016).

(36)	Mongolian (Öztürk 2013: 190)
	a.
	b.

Apart from the intonational break and the afterthought profile, Öztürk (2013) provides a variety of syntactic tests that argue against a mono-clausal movement analysis. Postverbal constituents are for example incompatible with idiom formation. In (37b), only the literal reading is available.

Moreover, corresponding gaps of the postverbal phrases in the antecedent clause can be pronominalized (38a), in clear contrast to leftward scrambling in Mongolian for which this strategy is not available (38b).

Given the discussion above, it can be concluded that postverbal constituents do not result from VP-movement. Rather, postverbal constituents are derived from a complex bi-clausal structure with subsequent deletion of all clausal material in the second clause except for the postverbal constituent.

As predicted, VPs do not undergo long scrambling either, see (39b).

Again, we are faced with the question of V-to-T movement. Fortunately, there is a class of adverbs in Mongolian with which we can test remnant VP-movement. Low manner adverbs like dandaa ‘always’ (Guntsetseg 2016: 25) and xurdan ‘quickly’ (Fong 2019: 16) cannot occur before the subject, see (40) for xurdan. Adverb movement is apparently not an option in Mongolian.

Such adverbs showing positional restriction can be used to test remnant VP-scrambling. Let us assume for now that V raises to T, thus vacating the VP, and low manner adverbs adjoin to VP. By dislocating the adverb together with the direct object, we can probe for VP-movement. The unacceptability of (41) provides evidence that VPs do not scramble across a subject. The adverb cannot precede the subject, not even if it is contained within VP.

The same pattern emerges in ditransitive scenarios, where (42a) and (42b) show that short as well as intermediate scrambling is not an option for remnant VPs.

Neither is long scrambling permitted with remnant VPs, exemplified with the paradigm in (43). The adverb sain ‘well’ is another low manner adverb which cannot precede the subject, shown in (43a). Scrambling the remnant VP across a sentence boundary is banned, (43b) is not acceptable.

Let us now turn to Tamil. Long scrambling of VPs seems to be unacceptable, judging by the contrasts in (44). Note, again, that tense and agreement morphology is affixed to the verb, suggesting potential V-to-T movement which would provide an alternative explanation for the unacceptability of (44b). VPs that contain inflected verbs would not be able to occur in dislocated position since head movement potentially bleeds VP-movement. In contrast to Mongolian, Tamil does not provide viable tests for VP-remnant movement.¹⁰ Hence, the VP-fronting facts are not entirely conclusive.

In order to test for VP-movement within clauses, we again consider the possibility of post-verbal constituents. As is shown in (45), subjects and indirect objects are principally able to occur in such positions. The translations, however, already signal a marked information structure with such word orders. They are often translated as clefts or pseudo-clefts in which the post-verbal phrase constitutes the pivot. Alternatively, they have been analyzed as mono-clausal structures with narrow focus on the post-verbal phrase.

Postverbal constituents in Dravidian and Tibeto-Burman languages are overwhelming classified as pivots for underlying clefts, not only because of their characteristic information structure but also due to the presence of an overtly spelled-out copula, often co-occuring to the right of the cleft clause (Annamalai & Steever 1998; Lehmann 1998; Krishnamurti 1998; 2003; Bhattacharya & Devi 2004). Comparable structures in Malayalam and Meiteilon are given in (46) and (47). Consequently, bi-clausal analyses are prevalent (Madhavan 1987; Jayaseelan 2001; Bhattacharya & Devi 2004; Jayaseelan & Amritavalli 2005; 2017; Selvanathan 2017).

Tamil, however, often does not spell-out the copula, thereby making it more difficult to diagnose a cleft. The examples in (48) prove that neither a predicative nor an identificational copula sentence requires an overt copula.

Sarma (1999; 2003) discusses two variants of Tamil post-verbal constituent structures, one in which the verb shows full agreement, see (49a) and (49b), and one with default agreement – or alternatively analyzed as a nominalizer (the two are homophonous in Tamil). The second variant is shown in (49c).

Let us first consider the default agreement/nominalizer version. These structures have received a bi-clausal treatment by Sarma (1999; 2003), and more recently Selvanathan (2017), in which the clefted phrase originates in the cleft clause but moves to a designated focus position. The accounts are presented in (50) and (51), applied to (49c). Sarma takes -adu to be the spell-out of N, a nominal layer above the cleft clause, shown in (50).¹¹

In contrast, Selvanathan analyzes -adu as the result of anti-agreement (Ouhalla 1993; Schneider-Zioga 2007) between an N-layer and the clefted phrase: Ø-agreement is disrupted because the clefted phrase is involved in another A’-dependency, here the Foc head.

The motivation for such analyses comes from case-connectivity and reflexive binding of the clefted phrase by an antecedent in the cleft clause, shown in (52). These observations were first made by Sarma (1999) and then later on also discussed by Selvanathan (2017).

Sarma (1999: 95) also provides an analysis for the full agreement versions in (49a) and (49b) which is identical to the structure in (50), minus the nominal layer and the copula phrase. Case connectivity also holds for full agreement structures, while (53) additionally shows reconstruction for reflexive binding.

Movement of the clefted phrase triggers weak crossover effects, shown by Sarma for the full agreement versions, see (54). Again, we can make a parallel observation for the default agreement/nominalizer versions, shown in (55).

While case-connectivity and reconstruction for Principle A serve as arguments that the clefted phrase undergoes movement out of the cleft clause, weak crossover effects indicate that it must be Ā-movement, which fits well with the information structural nature that is proposed under those accounts.

Note that neither of the analyses makes use of VP-movement. The observations can, however, be made to follow from such an analysis. Let us assume for the moment the analysis in (57) where the focused phrase vacates the VP into a leftward branching focus position, with subsequent leftward VP-movement across the focused phrase. This type of analysis has in fact been proposed for equivalent post-verbal structures in Malayalam (Jayaseelan 2001; 2004). Case can be assigned to the object in first-merged position, the same position into which the object reconstructs for reflexive binding. Weak crossover is not ameliorated since object movement is triggered by a focus feature, arguably an Ā-type movement. We can understand the analysis in (57) as the full agreement counterpart version of (51).¹²

There is, however, one piece of evidence suggesting that rightward movement for post-verbal constituents has to be at least an option. Tamil bans subextraction, that is extraction out of a phrase which has already been moved. This is shown with the set of examples in (59)–(61). An embedded CP can in principle move over the matrix subject, compare (59a) to (59b). In (60), we see that arguments can also dislocate to the right periphery across clause boundaries. Both movement operations, however, cannot co-occur, as the contrast in (61) exemplifies.¹³ Sarma attributes the unacceptability of (61b) to a Freezing effect (Ross 1967; Culicover & Wexler 1977). As in (59b), the embedded CP has been fronted. Crucially, further subextraction by rightward movement of the clefted phrase is not licensed.

If Tamil exhibits freezing effects, we should be able to create similar configurations in mono-clausal structures. In light of the discussion above and under the assumptions in (57), the missing contrast in (62) is unexpected. If postverbal constituents were to come about via remnant VP-movement, it should not be possible to subextract the wh-phrase from the VP in its derived position in (62b), nevertheless this seems possible.

The derivation of (62b) under the theory in (57) is sketched in (63).

An analysis that derives postverbal constituents directly via rightward movement makes the right prediction for (62) since the arguments undergo movement to the right and to the left periphery independent of each other. The derivation is given in (64).

This argument, of course, does not exclude the possibility of VP-movement altogether, but it does make it unlikely that VP-movement is responsible for post-verbal constituent structures in Tamil.

This concludes our presentation of Mongolian and Tamil – two PNI languages that maximally restrict the movement capabilities of PNI-ed objects. This section investigated the potential of VP-movement in the respective languages and found no evidence. Rather, post-verbal constituent structures most likely receive a bi-clausal analysis in Mongolian and Tamil, the former with subsequent deletion under identity, the latter in the form of a (pseudo-)cleft. The observations made in this section lend support to the idea that PNI-ed objects and VPs should be treated on par, in that both types of phrases are severely restricted in their movement capacities.

4.2 German & Turkish

German permits PNI-ed objects to undergo topicalization, i.e. movement to spec,CP, but not scrambling, a generalization that was established in section 3.3. As it turns out, VPs exhibit exactly the same movement restrictions as PNI-ed arguments. In contrast to Tamil and Mongolian, German does not necessarily fuse tense/agreement morphology with V. Hence, VP-movement can be diagnosed in a straightforward manner. It is widely acknowledged that German VPs can undergo topicalization, but not scrambling (Grewendorf & Sternefeld 1990; Grewendorf 1995; Müller 1998). A minimal pair is given in (65) with the underlying structures in (66) and (67).

The parallelism between movement of PNI-ed arguments and VP-movement provides an explanation for Frey’s puzzle in section 3.3. Recall that Frey (2015) assumes PNI to be complex predicate formation where fronting of the verbal part leads to unacceptability, but fronting of the nominal part does not. The minimal pair is repeated in (68).

In section 3.3, the unacceptability of (68b) was analyzed as a consequence of a ban on scrambling the PNI-ed object Karten out of the VP, so that remnant VP-fronting can take place. In (68a), the PNI-ed object topicalizes, thereby leaving the VP behind. We now see a clear parallel between licit topicalization in (68a) and (65a) and illicit scrambling in (68b) and (65b), providing further evidence for the parallelism between movement of PNI-ed arguments and VP-movement.

Considering the line of argumentation so far and the movement patterns presented in section 3.2, we expect VP-movement in Turkish to be freely available. This prediction seems to be borne out. The paradigm given in (69) shows that post-verbal constituents are acceptable in Turkish. Note, however, that these structures might as well be created via rightward movement into postverbal position, which has been shown to exist independently in Turkish (Kural 1997; Termücü 2005; Kornfilt 2005). As in Tamil and Mongolian, verbs are inflected for tense/aspect, suggestive of V-to-T movement. The size of the scrambled constituents in (70) might be even larger, considering the morphology signaling a nominalizer and case.

Since the VP-status of the moved phrases in (69) and (70) is not entirely conclusive,¹⁴ we test for remnant VP-movement, that is movement of the direct object together with a low manner adverb. The examples in (71) and (72) are in line with the previous observations, VPs are allowed to undergo long scrambling in Turkish.

Finally, let us look at an argument suggesting the independent need for VP-movement in Turkish. This argument is based on an observation coming from discontinuous possessor phrases. Turkish can extract possessors to the left and to the right periphery, shown in (73), but see also Bošković & şener (2014).

Displacement of the possessum, however, seems to be restricted, in that the possessum can only show up right peripherally, shown in (74).

Let us assume for now that the possessor is merged in the specifier position of a nominal projection that takes the possessum as a complement. We predict that the possessum can only move as a remnant DP, after the possessor has moved out. The structures in (74), thus, require two movement operations, one creating the remnant constituent (possessor extraction) and one which constitutes the remnant movement operation. If (74a) were derived via (i) rightward movement of the entire possessor phrase and (ii) subsequent subextraction of the possessor, sketched in (75), a freezing effect should be triggered, contrary to fact.

We would, however, make the right prediction for (74b) since possessor extraction targets a position not accessible anymore, after the possessor phrase has undergone long scrambling. This derivation is shown in (76).

A successful way to derive the acceptability of (74a), is to reverse the order of operations, i.e. (i) leftward extraction of the possessor and (ii) rightward remnant movement, possibly as adjunction to CP. This derivation is shown in (77).

The explanation for (74a), however, does not extend to (74b), which would be minimally different from (77) in that remnant movement occurs to the left, see (78), leaving us with no explanation for its unacceptability.¹⁵

We have, thus, arrived at an impasse. No possible order of operations is able to derive the contrast in (74). There is one option, however, we have not considered so far. Let us propose a general restriction on remnant possessor phrase movement in Turkish. This restrictions accounts for the unacceptability of (74b). The displacement of the possessum in (74a) must now be a consequence of VP-movement, with the derivation sketched in (79). Hence, in order to derive the acceptability of (74a), we crucially need VP-movement to be an option in Turkish.¹⁶

This section presented positive evidence for the parallelism between movement of PNI-ed arguments and VPs with respect to topicalization in German as well as scrambling in Turkish. Together with the previous section, we can come up with the cross-linguistic picture in Table 1.

The next section will lay out the analysis which essentially ties the parallelism observed in this section to the verbal nature of PNI-ed arguments.

5.1 Movement properties

The proposal is worked out in a minimalist framework (Chomsky 1995) – a derivational model of grammar in which the basic operations Merge and Agree apply in sequential order. Syntactic structures are build bottom-up by sequential application of Merge and Agree from a set of lexical items, taken from the numeration. Syntactic operations are driven by two types of features: (i) structure-building features [•F•] triggering Merge, where movement is defined as internal Merge, and (ii) probe features [*F*] triggering Agree. [•F•] and [*F*] must be targeted and discharged during the derivation, thereby restricting the possible outcome of syntactic derivations. If a head comes with more than one structure-building or probe feature, where for each feature the context to apply is met, the Earliness Principle (Pesetsky 1989) demands that the syntactic operations the features trigger apply either simultaneously or in a certain order. Recent analyses have made use of the latter option, deriving feeding and bleeding interactions between Merge and Agree, either implicitly (Anand & Nevins 2006; Asarina 2011) or explicitly by making reference to a feature list or feature stack (Stabler 1997; Müller 2009; 2010; 2011; Georgi 2014; Assmann et al. 2015; Heck & Himmelreich 2017). Following Müller (2011: 168), I assume that the features of a head constitute a list and will be discharged one after another, beginning with the first feature in the list. There are no syntactic operations which are not feature-triggered and every probe and structure-building feature can only be targeted once. They will be discharged, after they have undergone an operation, in order to make room for the next feature on the stack. Features become inactive after they have taken part in a structure-building or Agree relation, see also the discussion in Müller (2009: 288), Müller (2010: 40), and Georgi (2014: 109).¹⁸ Note that goal features do not have to be discharged for the derivation to converge. They will, however, nevertheless be discharged after they have taken part in an operation. Structural case is assigned by the functional heads T and v (Chomsky 1995; 2000) where a checking account of case assignment is adopted, in which both probe and goal enter the derivation with valued case features but Agree requires matching of features (Müller 2009; 2011).

PNI-ed arguments are the result of a special D head, a feature list that contains a categorial [D] feature as well as a categorial [V] feature. The features are ordered so that [D] is relevant for processes early in the derivation, i.e. c-selection and θ-role assignment, while [V] is at work for later operations such as realization of case features, scope, and movement. A sample derivation is given in (81)–(84), exemplarily shown for the PNI scenario in (2), repeated here in (80). The tree in (81) presents the internal feature structure of a PNI-ed argument, while (82) shows how a PNI-ed argument is selected for by the verb. Since [D] is ordered higher on the feature stack than [V], the PNI-ed object is c-selected like a proper argument.¹⁹ If we compare the feature stack of the subject with the one of the object in (84), we see that external Merge of each argument is triggered by a structure-building feature [•D•].²⁰ As is shown in (83), syntactic case assignment is not blocked in PNI contexts. Consequently, no other syntactic operation potentially interacting with case assignment should be affected by PNI-related case drop – a desired result since no interaction with Ø-agreement or valency reduction can be detected for the languages of this study.²¹ In section 5.2, a post-syntactic account of case drop is presented. PNI-ed arguments and proper arguments are of different categorial status once they are first-merged with their functional heads. This difference now accounts for the PNI effects investigated in this paper.

Having set-up the basic system, let us now turn to the movement patterns of PNI-ed arguments. Table 1 summarized the movement properties of PNI-ed objects within the languages investigated in this paper. All languages in this study allow for scrambling of nominal arguments. The overview signals that movement of PNI-ed arguments is generally restricted, yet not every language shows the same restrictions. An ideal way to account for the cross-linguistic variation is to adhere to individual properties of the languages under consideration. The present approach is able to do so in a straightforward manner. Since PNI-ed arguments constitute hybrid categories which start off as nominals but turn into verbal arguments, once they have been c-selected, we predict their movement patterns to parallel VP movement in the respective languages. Since VP-movement is often more restricted than DP-movement cross-linguistically, the movement patterns of PNI-ed arguments are not surprising.

Movement is driven by categorial features. While this is a commonly shared assumption for nominal arguments undergoing scrambling, EPP-movement, or object shift (Chomsky 1995; Kitahara 1997; Bailyn 2003 a.o.), category-driven movement triggers have also been proposed for VP/vP/PredP-movement (Massam & Smallwood 1997; Massam 2001; Mahajan 2003; Müller 2004; Collins 2017; van Urk 2019a).²² Hence, DP-movement is triggered by [•D•], whereas movement of VPs as well as PNI-ed arguments results from the presence of [•V•] on the respective functional heads in the clause, i.e. v for short scrambling, T for intermediate scrambling and C for long scrambling and/or topicalization. With these assumptions in place, I provide the featural set-up for the PNI languages under discussion in Table 2. In line with the distribution in Table 1, each allowed movement operation will be licensed by the presence of [•V•]. Turkish shows maximal flexibility considering PNI- and VP-movement since v and T can optionally come with the movement-inducing categorial feature [•V•]. By assumption, these options are not available to Mongolian and Tamil. Hence, PNI-ed arguments cannot leave their base positions. German allows for topicalization of PNI-ed arguments but blocks scrambling. Thus, only the C head can be equipped with [•V•].²³

How the movement patterns established in this study follow from the featural set-up in Table 2 will be demonstrated in the following, beginning with short scrambling. Indirect arguments are introduced by Appl which takes VP as a complement (Marantz 1993). Turkish PNI-ed objects undergo short scrambling if v comes with [•V•], thereby attracting the PNI-ed object to an inner specifier position, see (85a). The scrambling feature [•V•] is ordered before [•D•] which subsequently c-selects a proper DP, leading to external Merge of the subject as an outer specifier of vP, shown in (85b). The crucial features responsible for the indicated movement step are .

Mongolian, Tamil, and German do not exhibit short scrambling of PNI-ed objects because v does not come with [•V•]. Adding a nominal scrambling feature [•D•] does not help, as it would not target the right argument. Intermediate scrambling follows the same rationale, Turkish allows for PNI-ed objects to move across subjects since T comes with [•V•], shown in (86), whereas Mongolian, Tamil and German T do not have this option.

German permits topicalization of PNI-ed objects, enabled by a C head with a [•V•] feature, which is illustrated in (87).

Finally, long scrambling in Turkish is triggered by the scrambling feature [•V•] on matrix T, shown in (88). Since Mongolian, Tamil, and German cannot assign the scrambling feature [•V•] to T generally, they do not allow for long scrambling.

The analysis presented in this section can cover the cross-linguistic variation we find concerning movement restrictions of PNI-ed arguments. The parallelism with VP-movement is taken as evidence for the hybrid character of PNI-ed constituents. Since the current proposal ties movement restrictions to the verbal nature of PNI-ed arguments, a straightforward implementation suggests itself which makes reference to categorially triggered movement operations. The analysis is, in this sense, superior to traditional DP/NP approaches which essentially derive PNI properties from the lack of a DP/KP-shell, thus predicting complete immobility (Massam 2001; Dayal 2011; Barrie & Li 2015; López 2012; Frey 2015) or even surface adjacency (Baker 2014; Levin 2015). In the next section, the relevance of the partially verbal nature for two other core PNI properties will be motivated.

5.2 Extension to case and scope properties

Often implicitly assumed amongst many scholars is that case is uniformly expressed on nouns and not on verbs, see however Blake (2001/2004) and Moravcsik (2012) for explicit statements of such kind. This assumption finds empirical support in the work by Nichols (1986) who identifies case as the predominant morphological category for dependent marking strategies, whereas person, number, and gender morphology are most commonly expressed in head-marking patterns. An early implementation of this dichotomy can be found in the Principles and Parameters tradition of Generative Grammar. Lexical categories were distinguished by two binary distinctive features [±N] and [±V] where [–N] categories assign case and [+N] categories receive case (Chomsky 1981; Stowell 1981). The most recent installment of this assumption can be found within the theory of Dependent Case (Marantz 1991; Wunderlich 1997) where case marking expresses the licensing of one nominal in the local presence of another nominal. The lack of case-marking on PNI-ed arguments is thus a direct consequence of the non-nominal nature of PNI-ed arguments. To implement this idea in the current framework, I suggest a post-syntactic treatment of case drop, in line with many proposals for Differential Object Marking (Bossong 1991; Aissen 2003). Post-syntactic DOM/PNI accounts often make use of special spell-out rules (López 2012; Lidz 2006; Nuger 2010) or impoverishment rules (Keine 2007; Keine & Müller 2011; 2015; Weisser 2018) to account for case loss. Impoverishment rules (Bonet 1991) reduce morpho-syntactic feature bundles/lists by deleting (sub)features, thereby blocking insertion which inturn requires retreating to the general case, which is often an elsewhere marker that is spelled out as /∅/. Within the current system, impoverishment would be contextually triggered by [V] on PNI-ed arguments, deleting the case feature resulting in zero exponence. For theories that model case assignment in terms of Agree, this impoverishment rule must be in place regardless of PNI, as it explains why verbs rarely express case morphology. In contrast, non-incorporated arguments do not come with [V], which prevents them from undergoing case drop. An Optimality-based system such as the one proposed by Aissen (1999; 2003) can describe the case drop equally well by integrating PNI-ed arguments on the definiteness scale – one of the prominence scales assumed operative as primitives in grammar. Keine & Müller (2011; 2015) develop a post-syntactic version of Aissen’s OT-approach. Situating case loss in the post-syntactic component also accounts for PNI languages in which case marking is not affected. Languages such as German (Frey 2015) and Hungarian (Farkas & de Swart 2003) exemplify how PNI is not necessarily always characterized by a correlation between lack of case marking and scope inertness. This is entirely predicted by post-syntactic accounts to case loss. Impoverishment rules are motivated by markedness considerations but do not have to apply across the board. An OT-system can capture cross-linguistic variation in an equally adequate fashion by re-ranking the markedness constraint that leads to case drop.

Finally, let us address the low scope restriction. One of the main reasons why the syntactic status of head movement is so fiercly debated in the recent past comes from the observation that verb movement never seems to change scopal relations. This has lead many scholars, most prominently Chomsky (1995; 2000; 2001) with many following him (Boeckx & Stjepanović 2001; Merchant 2002; Harley 2004), to place head movement in general in the post-syntactic component. Others have argued for verb movement specifically to take place in syntax but with obligatory semantic reconstruction (Goldberg 2005; Matushansky 2006; Keine & Bhatt 2016). The scopal properties of PNI-ed arguments thus parallel those of verbs, in that they cannot take scope over another operator in the sentence. Under the assumption that syntactic categories are related to semantic types in a meaningful way (von Stechow 2012), I assume that [V]-features are intrinsically linked to events, whereas e.g. nominals are linked to individuals, CPs are linked to worlds etc. A promising way to restrict the scope of PNI-ed arguments is to enable interpretation only in the event domain. With Kratzer (1995) and Chung & Ladusaw (2004), I assume that verbs introduce an event variable ⟨v⟩ which is accessible within the verbal domain but not outside of it, as it is closed off immediately above the verbal domain (vP). The PNI determiner is an existential quantifier of type ⟨⟨e,t⟩,⟨e,⟨v,t⟩⟩⟩, thus it can be interpreted only in the verbal domain. Generalized quantifiers such as the universal quantifiers in (1)–(3) are of type ⟨⟨e,t⟩,t⟩ and have to be interpreted outside of the event domain (Landman 2000), thereby deriving the scope effects in (1)–(3). An elaborate discussion on the case and scope properties can be found in Driemel (2020a; b).