German allows the inversion of possessive PPs, as seen in (1a). It has previously been observed by Gallmann & Lindauer (1994) that this inversion is banned inside the complement of PPs (see (1b)). We collected reliable data that show that this inversion is ungrammatical in dative contexts, too (see (2)). These restrictions contrast prepositional possessors with other kinds of possessors in German which can be inversed in these contexts (see (1c) or (2c)).

In the following paper, we will show that this restriction follows directly in an analysis based on feature-driven Merge in a Minimalist framework. As the wide-spread constraint Merge-before-Move (Castillo et al. 1999; Frampton & Gutmann 1999; Chomsky 2000; Müller & Sternefeld; Hornstein 2001; 2009; Boeckx et al. 2010; Drummond 2011; Weisser 2015) enforces the higher P head to check the structure-buildung [•P•] feature, movement of the possessive PP is blocked. This paper is structured as follows. The relevant data is presented in Section 2. In Section 3, we propose a new analysis. Before concluding in Section 5, we discuss further issues of our proposal in Section 4.

In German, there are four different constructions that encode possession (Georgi & Salzmann 2011): The possessor may appear in genitive case, as in (3a). Furthermore, there is the prenominal genitive construction in (3b) and the double possessive construction in (3c). In this paper, however, we will focus on the construction in (3d) including a possessor preceded by the preposition von (‘of’).

In non-standard German, the construction allows possessee and possessive PP to swap their positions, as seen in (4). The order of possessee and von-possessor in (4a) is the standard word order, which is widely accepted among German native speakers. In (4b), however, the constituents appear in reverse order.

Gallmann & Lindauer (1994) observed that the inversion cannot take place if the DP is inside a PP, as seen in (5b).1

A small-scale grammaticality judgment task in which we asked 49 native speakers of German to rate sentences from 1 (perfectly grammatical) to 6 (ungrammatical) showed that the inverse order is ungrammatical in dative contexts, too. While the construction allows the possessee to bear accusative case (see (6a)), dative possessees lead to ungrammaticality, as seen in (6b)–(6d). Note that this restriction is independent of the position of the constituent (prefield in (6c) vs. midfield in (6d)) or the status of the dative (lexical dative in (6b) vs. structural dative in (6c)).23

An anonymous reviewer pointed out that the grammaticality of (6b) increases when Hans is heavily stressed. Assuming that stress correlates with focus, we follow Fanselow (2002); Frey (2006) and Frey (2010) that the stressed PP has been moved to a designated focus position. Thus, the example in (7a) is structurally different from the example in (6b) as the possessive PP in (7a) has been scrambled out of the DP giving rise to the grammaticality of split DPs in (7b). In contrast, the possessor in (6b) in SpecDP remains in the DP.4

Moreover, the following examples show that other possessive constructions are not subject to these blocking effects and may be used with datives (see (8)) and after prepositions ((9)).

From the data we have seen in this section, we can draw the generalization that the the inversion of prepositional modifiers is ungrammatical PP-internally, as previously observed by Gallmann & Lindauer (1994), as well as in dative contexts. As already pointed out by Gallmann & Lindauer (1994), these restrictions are not captured by previous analyses of this phenomenon, such as Bhatt (1990) or Fortmann (1996). Even though they suggest movement analyses within a Government and Binding framework, they fail to derive the blocking effect in the complement of PPs since the trigger for movement within G&B does not arise from features but from the relation of a moved XP and its trace. Hence, the goal of this paper is to present an analysis of German PP-inversion that can account for the restrictions that this construction is imposed on. In the following section, we will show that our proposal analyzes the ban on PP-internal PP-movement as an instance of a more general constraint on movement, the Internal Inversion Constraint, which arises from the constraint known as Merge-before-Move.

For the following analysis, we will adopt a Minimalist framework with feature-driven Merge, i.e. Merge and Move (or: base-Merge and re-Merge) can only take place if they are triggered by a structure-building feature [•X•] (Heck & Müller 2003; 2007). This assumption is based on the last resort condition given in (10).

Another assumption made her, is that the inversion is triggered by an optional feature [•P•] which is the lowest feature in an ordered feature bundle as proposed by Georgi & Müller (2010). In this analysis, this feature bundle is on D, such that D has the featural specification [•N• ≻ •P•].5 The feature introduces a prepositional modifier and triggers movement of a PP to SpecDP. If the DP is not embeddded in a PP, the derivation thus proceeds as follows. In the first step, D satisfies its [•N•] feature by merging with the NP. At this point of the derivation (see (6-a)), the remaining [•P•] feature on D has not been satisfied. This can now be achieved ONLY by moving the possesssive PP to SpecDP, as seen in (11).

In the next step, we will see that this inversion cannot take place in the complement of PPs. Let us assume that we are at the point of the derivation where a complex NP has been merged with D. The next feature to be satisfied is the [•P•] feature on D. If the next head in the numeration is also a P head, [•P•] on D can be satisfied either by moving the possessive PP to SpecDP or by merging the next head in the numeration.6 Due to the constraint Merge-before-Move in (12) (Castillo et al. 1999; Frampton & Gutmann 1999; Chomsky 2000; Müller & Sternefeld 2001; Hornstein 2001; 2009; Boeckx et al. 2010; Drummond 2011; Weisser 2015), the next head of the numeration is merged which blocks the movement of the possessive PP to SpecDP as the movement trigger has already been checked.7

The constraint on movement that is responsible for the PP-internal blocking effect is summarized in (14) as the Internal Inversion Constraint. This constraint does not need to be stipulated, since it follows from the more general Merge-over-Move constraint.

Meinunger (2000); Bayer et al. (2001); McFadden (2004); Rezac (2008), among others, have previously shown that datives and PPs form a natural class which has lead to the assumption that datives are actually PPs with a silent P head. Bayer et al. (2001), for instance, argue that all datives in German are actually embedded in a larger KasP which may be headed by a silent element or an overt P. One of their arguments comes from topic drop, where neither datives nor PPs can be dropped while nominatives and accusatives can, as seen in (15).

We have seen that our approach makes reference to categorial features, in contrast to previous analyses by Bhatt (1990); Fortmann (1996) and Georgi & Müller (2010). Hence, our analysis automatically covers the fact that prepositional modifiers show case restrictions (see the data (16), repeated from Section 2) under the assumption that datives are actually PPs. This being the case, it is logical consequence that both datives and PPs are subject to the Internal Inversion Constraint.

Moreover, the fact that the Internal Inversion Constraint is sensitive to categorial features predicts that the blocking effect should arise not only with possessives PPs but with other prepositional modifiers, as well. As seen in (17), this prediction is borne out.

In this section, we have seen that the restrictions of prenominal possessive PPs follow automatically from a [•P•] feature on D and the constraint Merge-before-Move. Moreover, it can easily be explained why only prepositional modifiers are subject to this restriction. The competition between the embedded PP and the next head in the numeration only arises as they are of the same lexical category. Since other kinds of possessors are non-prepositional, no blocking effect is expected.

In this section, we will discuss some open questions and challenges for our account, before providing two putative examples from Archi nominalizations and Swiss German embedded V1 clauses. These could be seen as more general consequences of the proposed Internal Inversion Constraint.

At this point, we want to discuss a rather unconventional issue of the otherwise simple account that we put forward in this paper. As the D head bears a [•P•] feature and the next higher P head bears a [•D•] feature, our approach presupposes reciprocal subcategorization. This automatically leads to the question which of the two heads projects. Chomsky (2013) proposes a Labeling algorithm (see (18)) that allows a head to project if it merges with a phrase. In his view, a head is not defined as a constituent that subcategorizes but negatively as a syntactic object that is not the result of Merge. In our analysis, it thus follows that P would still be the projecting head which clearly solves the problem of reciprocal subcategorization.8

In section 3, we proposed the Internal Inversion Constraint that follows directly from the basic assumptions of feature-driven syntax and Merge-before-Move. If the Internal Inversion Constraint holds for all categories, we should expect similar blocking phenomena in other languages. Even though the distribution of similar blocking effects exceeds the scope of this article, we want to mention two possible instances of the Internal Inversion Constraint in other languages.

As Lötscher (1997) and Dürscheid & Hefti (2006) point out, one remarkable property of Swiss German syntax are embedded V1-clauses usually being headed by predicative adjectives or nouns (see (19)). The frequency of V1-clauses headed by verbs remains unclear in the literature, however, Lötscher (1997) mentions that the acceptability of V1-clauses strongly increases if they are headed by adjectives. This can be explained by means of the Internal Inversion Constraint such that V-to-C movement is blocked in clauses that are embedded under verbs but not under adjectives and nouns.

Another possible instance of the Internal Inversion Constraint is found in Archi (Nakh-Dagestanian, Russia). In this language, scrambling is allowed in main clauses, but not in nominalized clauses (Bond et al. 2016).9 They note that in nominalized clauses “[t]he genitive possessor […] can appear only in the very beginning or the very end of the clause, but not between the absolutive and the verb (whereas the ergative can take this position).” (Bond et al. 2016, 66). This blocking effect can be explained in terms of the Internal Inversion Constraint under the assumption that nominalized clauses are headed by a silent D head which blocks movement of a DP to a higher position.

In this paper, we have clarified and extended the empirical base of prepositional possessors in German. We have proposed an analysis of the inversion of prepositional modifiers in German. We have shown that the restriction against PP-internal PP-inversion follows from two basic assumptions of feature-driven syntax, namely the Last Resort condition and Merge-before-Move. If a head in the numeration and a lower phrase in the structure compete to check a structure-building feature, Merge will always be preferred, which leads to a blocking effect. This effect can be generalized as the Internal Inversion Constraint which makes clear and falsifiable predictions for other domains in syntax.