Jump to ContentJump to Main Navigation
Scrambling, Remnant Movement, and Restructuring in West Germanic$

Roland Hinterhölzl

Print publication date: 2006

Print ISBN-13: 9780195308211

Published to Oxford Scholarship Online: January 2010

DOI: 10.1093/acprof:oso/9780195308211.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 17 September 2019



(p.3) 1 Introduction
Scrambling, Remnant Movement, and Restructuring in West Germanic

Roland Hinterhölzl (Contributor Webpage)

Oxford University Press

Abstract and Keywords

This chapter begins by examining the connections between remnant movement, scrambling, and restructuring. It introduces a number of phenomena and concepts essential for the description of the syntactic structure of West Germanic as well as for the understanding of the discussion of restructuring infinitives. It discusses the essential properties of and the relevant restrictions on remnant movement. It also talks about the interaction between remnant movement and head movement. It argues that remnant categories created by head movement cannot undergo further movement and show how this restriction can be derived from Attract Closest as well. It demonstrates that the original account by Den Besten and Webelhuth is flawed. It argues that remnant VPs in German are created by licensing movement of VP-internal material into dedicated licensing positions in the lower middle field. It outlines the core ideas and concepts that were adopted and indicates the account of individual phenomena.

Keywords:   remnant movement, scrambling, restructuring, West Germanic, infinitives, head movement, Den Besten, Webelhuth

T his book is dedicated to an exploration of the empirical properties and the theoretical implications of three salient phenomena of West Germanic, namely, remnant movement, scrambling, and restructuring. The following section provides a brief outline of the interdependence of these phenomena.

1.1 Basic observations

1.1.1 The connection between remnant movement and scrambling

The term remnant movement was coined by Den Besten and Webelhuth (1987) to account for a peculiarity of VP-fronting in German. Consider example (1). It seems that what has been fronted (into [Spec,CP]) is just the verbal head. In order to reconcile this fact with the well-known Verb Second Constraint, which requires that the finite verb in matrix clauses is preceded by one (and only one) maximal phrase, Den Besten and Webelhuth propose to analyze the structure in (1) parallel to the case in (2), where the entire VP has been fronted.


gelesen hat Hans das Buch


read has Hans the book


das Buch gelesen hat Hans


the book read has Hans

(p.4) What makes (1) different from (2) in their analysis is the fact that in (1) the direct object has scrambled out of the VP prior to VP-fronting. The subsequent operation of VP-fronting will then move whatever remains in the VP, the so-called remnant category, into [Spec,CP], as is illustrated in (3). Thus, it seems that remnant movement of the VP is dependent on scrambling, which evacuates the VP of its internal argument. Since English does not allow scrambling, the parallel structure to (1) in English is correctly ruled out. If the VP is fronted in English, the object has to move along with the verb (3b–c).



[CP[VP tSCR gelesen] hat [IP Hans das BuchSCR tVP]]



John promised to read something



*… and read he did the book



John promised to read the book



… and read the book he did

1.1.2 The connection between remnant movement and restructuring

In German, VP-fronting is not restricted to participles but may also affect infinitives. Example (4a–b) shows that VP-fronting can apply to bare infinitives as well as to to-infinitives. However, there is one restriction to this operation. VP-fronting may affect an infinitive only if the matrix verb belongs to the class of restructuring verbs. This is illustrated by the contrast in (4c–d), where the matrix verb try, a typical restructuring verb, licenses fronting of the embedded VP, while the verb deny, a typical representative of non-restructuring verbs, fails to license the respective operation. Thus, it seems that remnant VP-fronting is dependent not only on scrambling but also on restructuring.



kaufen will er das Buch



buy wants he the book



zu kaufen versprach er das Buch



to buy promised he the book



zu kaufen hat er das Buch versucht



to buy has he the book tried



*zu kaufen hat er das Buch abgelehnt



to buy has he the book refused

1.1.3 The connection between scrambling and restructuring

Scrambling in German may permute the arguments of the verb. For instance, in (5a) the direct object has scrambled across the subject. It is important to note that this operation is clause-bound. As (5b) shows, the embedded object may not scramble (p.5) across the matrix subject. As (5c) shows, this also holds for embedded nonfinite clauses. However, in restructuring contexts, the embedded object can do exactly that, namely, scramble up to a position above the matrix subject, as is illustrated in (5d).



weil den Max jeder kennt



since the Max-ACC everyone-NOM knows



*well den Max Peter glaubt dass jeder kennt



since the Max-ACC the Peter-NOM believes that everyone-NOM knows



*weil den Max jeder zu kennen bedauert



since the Max-ACC everyone-NOM to know regrets



weil den Max jeder zu kennen glaubt



since the Max-ACC everyone-NOM to know believes

Thus, scrambling has often been used as a diagnostic for detecting restructuring infinitives. At this point we hit a bifurcation and two ways of explaining the contrast in (5c–d) seem to suggest themselves. First, one may hold that scrambling is always strictly clause-bound, implying that restructuring constructions are monoclausal structures (in this sense, scrambling is a diagnostics for the clause-mateness in infinitival constructions). Under these assumptions, studying restructuring constructions would amount to deriving a theory of which (verbal) predicates may combine within a single clause (cf. Cinque [2001], Wurmbrand [2001]).

Alternatively, one may assume that restructuring constructions are biclausal. Under this assumption, studying restructuring involves investigating the conditions under which an element may scramble from one clause into the next one up—in other words, under which conditions an element may undergo so-called long-distance scrambling. In this sense, scrambling is a diagnostics for the transparency of infinitival constructions.

Once the direct object of the embedded infinitive has undergone scrambling, the remnant infinitival phrase, however large it may be, probably just a VP in the mono-clausal approach and possibly an entire CP in the biclausal approach, is free to undergo fronting, providing a possible account for the contrast in (4c–d), as is illustrated in (6).



[hat er versucht [das Buch zu verkaufen]]

base structure



[hat er [das Buch]SCR versucht [tSCR zu verkaufen]]

long-distance scrambling



[[tSCR zu verkaufen] hat er das Buch versucht]

fronting of the infinitival

In this book, I will carefully investigate the phenomenon that has been called scrambling (chapter 2) and argue that scrambling is a cover term for various operations with different properties and triggers. It will be shown that one of these operations, so-called S-scrambling, is actually not clause-bound, but that all other scrambling operations are strictly clause-bound. Furthermore, I will argue on the basis of a thorough analysis of remnant movement and its restrictions (see section 1.4. later) that Den Besten (p.6) and Webelhuth’s famous account of data like (1) is untenable, since scrambling cannot be taken to create the remnant category that is the input for VP-fronting.

In (4c–d) earlier, I have illustrated that VP-fronting, an operation of remnant movement, is dependent on restructuring. In turn, I will argue that restructuring is dependent on remnant movement. The goal of this book is to argue that restructuring is based on remnant movement of large projections, both of the argument domain and of the licensing domain, of the infinitival clause (chapter 4). More specifically, I will show that S-scrambling does not play a role in restructuring. Thus, accounts of restructuring that assume that the constituents of the embedded infinitive undergo long-distance movement into the matrix clause are doomed. I will also show that what has been fronted in (4c) is not an infinitival clause (out of which the direct object has been scrambled into the matrix clause) but constitutes just an infinitival VP.

These observations, which I consider the empirical core of the book, seem to support a monoclausal account of restructuring constructions. However, I will argue (in chapter 5) that only a biclausal approach can provide a comprehensive and coherent account of all cases of restructuring in German, making necessary an account of restructuring in terms of remnant movement that splits up the embedded clause in various parts and has them licensed in dedicated positions in the matrix clause.

Before we take a closer look at restructuring, I have to introduce a number of phenomena and concepts that are essential for the description of the syntactic structure of West Germanic as well as for the understanding of the discussion of restructuring infinitives.

1.2 The basic syntactic phenomena of West Germanic

In this section, I provide an informal account of the basic notions and processes that govern the syntax of West Germanic. Though the different phenomena are introduced and illustrated by German data, the relevant generalizations carry over to the other West Germanic languages and dialects.

1.2.1 Verb second

First, note that German is a so-called Verb Second language. This means that the finite verb in matrix clauses moves from its base position, which is generally taken to be the position it occupies in an embedded clause (7a), to a sentence-initial position that has generally been assumed to be C0 (Den Besten 1977/1983). This movement only affects the finite part of a verb or verbal complex, leaving behind nonfinite verbs as in (7b) or verbal particles if they are separable (cf. [7c], which contains the particle verb mit-teilen). German is called Verb Second (V2) because in many sentence types the verb in this position must be preceded by another constituent, which is generally analyzed as occupying [Spec,CP] (7b).



weil Hans der Maria das Buch gegeben hat



since Hans to Mary the book given has



*(Gestern) hat Hans der Maria das Buch gegeben [t]



yesterday has Hans to Mary the book given

(p.7) The V2-structure in (7b) neatly displays the three topological fields, the Vorfeld (prefield), Mittelfeld (middle field), and Nachfeld (postfield), that have become standard terminology in describing the syntactic structure of German sentences. A constituent that precedes the finite verb in C0 occupies the Vorfeld or is said to be topicalized. Elements that occur between the finite verb and the nonfinite verb (or sometimes the verb-particle in case there is no auxiliary) are said to make up the middle field of the German sentence. A constituent that occurs after the nonfinite verb (or a verb-particle) occupies the Nachfeld and is assumed to be extraposed. For instance the subject Hans is topicalized in (7c), the sequence der Maria gestern forms the middle field of the sentence (7c), and the sentential complement dass sie krank ist is extraposed.



Hans teilte der Maria gestern mit [t] dass sie krank ist



Hans shared Mary-DAT yesterday with that she sick is



‘Hans shared with Mary (told Mary) that she is sick’

1.2.2 Scrambling and topicalization

Second, I want to introduce the notion of scrambling in an informal way. Chapter 2 provides an extensive discussion of the formal nature of scrambling; here it suffices to characterize scrambling as a movement process that allows for the re-arrangement of constituents in the middle field, as is illustrated in (10) later. This notion of scrambling presupposes that there is an (unmarked) basic word order, from which alternative orders are derived via scrambling and that this base order is identifiable with independent criteria (cf. Höhle [1982], Lenerz [1977], for some discussion of these issues). One of the tests for identifying base orders is the naturalness of certain orders in out-of-the-blue contexts, as is illustrated in (8).


Was ist passiert? Ich habe gehört


What happened? I have heard


a. dass der Hans die Maria getroffen hat



that Hans-NOM Maria-ACC met has


b. dass die Maria der Hans getroffen hat



that Maria-ACC Hans-NOM met has

The order in (8a) is unmarked in this context and thus counts as base order. The order in (8b) is inappropriate in this context and thus counts as a marked word order that is derived from the basic word order. This test reveals that basic word orders in German are verb class–dependent (cf. Haider and Rosengren [1998]). This means that the base order cannot simply be determined by the Case properties of the arguments of the verb, say NOM > DAT > ACC, but is dependent on the specific (thematic) class a verb belongs to. These classes are illustrated in the following table, adopted from Haider and Rosengren (1998).




Nom > Acc: bedauern ‘regret’, interpretieren ‘interpret’



Acc > Nom: interessieren ‘interest’, imponieren ‘impress’



Nom > Dat: helfen ‘help’, gratulieren ‘congratulate’



Dat > Nom: gefallen ‘appeal/please’, fehlen ‘lack’



Nom > Dat > Acc: anvertrauen ‘entrust’, verbieten ‘forbid’



Nom > Acc > Dat: aussetzen ‘expose’, unterordnen ‘subordinate’

Returning to scrambling and assuming that (10a) exemplifies the basic word order, scrambling can be described as a process that moves a constituent (preferably an argument) across arguments and adverbs to the top of the middle field. For instance, in (10b) the subject has been scrambled across both adverbs. In (10c), in addition to the scrambling in (10b), the direct object has been scrambled across the indirect object and the lower adverb, and so on. As should be clear from Example (10), the alternations I list only display a small subset of all the scrambling possibilities: In (10) I only gave the permutations that have the subject in the highest position in the middle field, then there are the permutations that have the subject in second position in the middle field and so on, not all of which, but most of which, yield felicitous sentences in one context or other.



weil vermutlich gestern Hans der Maria das Buch gegeben hat



since presumably yesterday Hans to Maria the book given has



weil Hans vermutlich gestern der Maria das Buch gegeben hat



weil Hans vermutlich das Buch gestern der Maria gegeben hat



weil Hans vermutlich gestern das Buch der Maria gegeben hat



weil Hans vermutlich der Maria das Buch gestern gegeben hat



weil Hans der Maria das Buch vermutlich gestern gegeben hat

For our purposes, it is irrelevant at this point what motivates scrambling or which alternations are most felicitous in which contexts. It is also irrelevant at this point whether scrambling is A- or A’-movement and how it is formally to be distinguished from topicalization. What is important for us here is that we assume that the alternations in (10) are derived by movement that we call scrambling and do not constitute different patterns that can be base generated, as has been proposed repeatedly by different scholars such as Fanselow (2001), Haider (1994), and Neeleman (1994a). To summarize, we characterize movement into the middle field as scrambling and movement into [Spec,CP] as topicalization.

It is important to note that this type of scrambling is clause-bound. Example (11) shows, while there is long-distance topicalization (11a), there is no process of long-distance scrambling in German (11b). In (11a), the embedded object has been moved into the topic position of the matrix clause and the result is fine. Scrambling (p.9) of the embedded object into the middle field of the matrix clause, however, results in ungrammaticality.



Den Mann hat Peter geglaubt, dass Maria t geküsst hat



the man-ACC has Peter believed that Mary kissed has



‘It was the man who Peter believed that Mary kissed’



*Peter hat den Mann geglaubt, dass Maria t geküsst hat



Peter has the man-ACC believed that Mary kissed has

Scrambling and topicalization also differ in the following respect: Scrambling seems to be restricted to arguments, while almost any (maximal) category, with the exception of IP, can be topicalized. Predicative elements, like predicate nominals (12a), heads of small clauses (12c) and idiomatic expressions (12e) cannot be scrambled, while the same elements can be topicalized quite naturally (cf. [12b, d, f]). Even verb-particles when they are separable from the verb, as can be detected from their behavior under V2 (whether they go along with the verb or stay behind in the base position), may be topicalized if the contribution of the particle to the meaning of the complex verb is transparent, while scrambling of verb-particles results in strong ungrammaticality (cf. [13b–c]). The particle weg (‘away’) is a separable verb prefix as can be seen from the alternation in (13a): When the main verb is finite it undergoes V2 and leaves the particle behind.



??Er hat einen Idioten gestern seinen Freund genannt



he has an idiot-ACC yesterday his friend called



‘He called his friend an idiot yesterday’



Einen Idioten hat er gestern seinen Freund genannt



an idiot has he yesterday his friend called



??Er hat grün gestern den Zaun gestrichen



he has green yesterday the fence painted



‘He painted the fence green yesterday’



Grün hat er gestern den Zaun gestrichen



green has he yesterday the fence painted



*Sie ist ihm ins Wort noch nie gefallen



she is him into-the word yet never fallen



‘So far she has never interrupted him’



Ins Wort ist sie ihm noch nie gefallen



into-the word is she him yet never fallen



Er ist noch nie weggelaufen/Er lief noch nie weg



he is not yet away-run/He ran not yet away



Weg ist er gelaufen (und ist nie wieder zurückgekommen)



away is he run (und is never again back-come)



‘He ran away (and has never come back again)’



??Er ist weg noch nie gelaufen



he is away not yet run



‘So far he has never run away’



*Gelaufen ist er noch nie weg



run is he yet never away

(p.10) It should be pointed out that the topicalizability of a separable prefix (particle) is a bit surprising. When the finite verb undergoes V2 to C0, [Spec,CP] has to be filled. X-bar theory prescribes that it be filled with a maximal projection. What seems to have been topicalized in (13b) is a head or even part of a head, if we analyze the otherwise inseparable string weg-laufen as a complex head. However, if we assume that a separable prefix starts out as an independent syntactic unit that heads its own projection that can either cliticize onto the verb by head movement or move as a maximal projection (when it receives some kind of contrastive stress), then the fact in (13b) can be brought in line with standard assumptions about topicalization. Finally, it should be noted that a particle verb cannot be topicalized leaving behind its separable prefix, as is shown in (13d).

1.2.3 Extraposition

It is important to note that the term extraposition, though we intend to use it as a pretheoretic notion to refer to constituents that occupy the postfield, has been used to describe a movement operation in what I will call the standard theory of the phrase structure of the West Germanic languages. The standard theory assumes that these languages are head-final and analyzes constituents to the right of the nonfinite verb as being extraposed, that is, moved to this position by right-adjunction to VP or IP from its base position that precedes the verb that governs it. Example (14a) shows the base structure, where a complex sentential complement precedes the matrix verb that governs it. In (14b), the sentential complement of the matrix verb has been extra-posed with the lower infinitival complement remaining in its (preverbal) base position. In (14c), both the complement of the matrix verb and the complement of the infinitival have been extraposed, that is, right-adjoined to the immediately dominating VP/IP note. It is important to note that extraposition is clause-bound. In (14d), the lower infinitival clause is moved across the dominating infinitival clause and right-adjoined to the matrix VP/IP, resulting in ungrammaticality. Viewing extraposition as derived by movement, we may assume that this operation may not cross a CP-node.1



dass der Lehrer [VP die Kinder [CP [CP die Aufgaben zu lösen] zu versuchen]






that the teacher the children the problems to solve to try encourages



‘that the teacher encourages the children to try to solve the problems’



dass der Lehrer [VP [VP die Kinder t bestärkt] [CP die Aufgaben zu lösen zu






dass der Lehrer [VP [VP die Kinder t1 bestärkt] [CP1 [VP [VP t2 zu versuchen] [CP2 die



Aufgaben zu lösen]]]]



*dass der Lehrer [VP [VP die Kinder [CP t zu versuchen] bestärkt] [die Aufgaben zu




(p.11) 1.2.4 Verb-finalness

As we have seen earlier, in embedded clauses the finite verb occurs in a sentence-final position (15a). Nonfinite verbs follow their complements (15b) and generally precede the finite verb in clause-final position (15c).



weil Hans gestern das Buch las



since Hans yesterday the book read



Hans hat gestern das Buch gelesen



Hans has yesterday the book read



weil Hans gestern das Buch gelesen hat



since Hans yesterday the book read has

When we compare this state of affairs with the positioning of finite and nonfinite verbs in a VO language like English, there are two options to account for these differences. The first option consists in assuming parametric variation in the head complement order and represents the standard approach to the syntax of the West Germanic OV languages. In this approach, the lexical and functional projections in the IP-domain are assumed to be head-final. Thus, the embedded verb-final position in (15a) is compatible either with the lack of V-movement (16a) or with rightward V-movement to a functional head within IP (16b). Furthermore, because of the right-headedness of the VP, an argument that precedes the verb may be analyzed as occupying its base position in the VP or as having scrambled out of the VP to a position in the middle field (cf. [16b] and [16c]). A considerable amount of work in German linguistics, culminating in Diesing’s (1992) mapping hypothesis, was dedicated to the question of which factors determine whether scrambling out of the VP does apply or does not apply.



[CP weil [IP Hans gestern [VP das Buch las]]]



[CP weil [IP Hans gestern [VP das Buch ti] lasi ]]



[CP weil [IP Hans gestern [das Buch]j [VP tj ti] lasi]]



since Hans yesterday the book read

The second option is to adopt the Universal Base Hypothesis (Kayne 1994), according to which all phrase structure is head-initial. In this approach it is assumed (p.12) that complements in German like in English are base generated following the selecting head and move to a licensing position to the left of the selecting head in the course of the derivation (cf. Zwart [1993]). The derivation of (15a) is illustrated in (17). In this approach, licensing movement is taken to derive the unmarked or neutral word order (in this case, the direct object moves to an Agreement head to check its Case, as I will argue later), before scrambling may further affect constituents in the middle field to derive more specialized or marked word orders. Therefore, this account predicts that scrambling is to be distinguished from movement out of the VP. I will take up this issue in section 1.4. on remnant movement.



[CP weil [IP Hans gestern [VP las das Buch]]]



since Hans yesterday read the book



[CP weil [IP Hans gestern [das Buch]j [VP las tj]]]



since Hans yesterday the book read

Note that in this approach, leftward movement must not only affect nominal arguments but also all VP-internal (verbal and non-verbal) predicates. In the case at hand, the participle in (15b–c) must move leftward to derive the unmarked word order PART > AUX.

A number of questions arise in this approach: First there is the question of where the finite verb in embedded clauses is located. It could be in its base position in the VP, as is assumed in (17b), or in a sentence medial I-position, as has been argued for by Haegeman (2001, 2002b). In the latter approach, movement of the finite verb to a medial Tense-head is followed by remnant movement of the VP to an even higher position in the middle field to derive the basic order arguments > verb. Throughout this book, I will adopt the former approach and assume that the verb in embedded clauses stays in the VP and can maximally move to a low functional position, below T (called Aspect, which will be introduced and motivated later). Second, the question arises of what the positions are to which VP-internal arguments and predicates (and VP-remnants in the account of Haegeman) move. And most important, what are the motivations and triggers for all these movements?

In the remainder of the book, I will try to provide satisfactory answers to these questions by investigating the properties of verbal complexes (chapter 3 and 4) as well as the distribution of arguments and adjuncts in the middle field (chapter 2). The general procedure will be the following: I will first adopt the standard approach to restructuring, scrambling, and remnant topicalization and point out the difficulties it faces in accounting for these phenomena. In exploring the alternative approach, I will provide empirical evidence for leftward movement of VP-internal material and develop an antisymmetric account of the syntactic core phenomena of the West Germanic OV languages.

In the process of developing the Universal Base approach, I will often provide analyses of phenomena in both accounts, switching back and forth between them in order to evaluate their advantages and disadvantages. Thus, it is important (p.13) for the reader to keep in mind the basic tenets of the two approaches as outlined earlier.

1.2.5 Comparing the standard approach and the Ubh approach

To summarize, in the standard approach the West Germanic OV languages are analyzed as head-final. The following processes are relevant for the description of a sentence in West Germanic: Verb Second moves the finite verb into C0 in matrix clauses (and certain complementizerless embedded clauses). Topicalization moves a constituent into the Vorfeld, which is identified as [Spec,CP]. Scrambling is an operation that allows for the rearrangement of syntactic constituents in the middle field. In the standard approach, scrambling is analyzed as an operation that moves constituents out of the VP. Extraposition moves a constituent into the Nachfeld. The core candidates for extraposition are embedded clauses. These elements are analyzed as being right-adjoined to VP or IP.

In the approach that endorses the UBH, the West Germanic OV languages are analyzed as head-initial languages, in which VP-internal material, both arguments and predicates, is moved leftward into licensing positions in the middle field (but see Haegeman 2001 and 2002b for an interesting alternative). Scrambling in this approach must be distinguished from licensing movement. The latter moves material out of the VP to derive the unmarked word order, while the former affects constituents in the unmarked word order to derive more specialized or marked word orders. Extraposition and rightward movement in general are disallowed. While in the standard approach the finite verb in embedded clauses may be taken to stay in the VP or to move to a clause final I-position, I will be adopting an antisymmetric approach in which the finite verb stays low and does not move to the clause-medial I-position.


(18) Standard Approach



(19) UBH Approach

Having familiarized ourselves with the basic tenets of the syntactic structure of German, we are now able to discuss (and appreciate) the properties that distinguish restructuring from non-restructuring infinitives.

1.3 Coherent infinitives

The West Germanic SOV languages German, Dutch, Frisian, and their numerous dialects have two types of infinitival complements. Like English, these languages have non-coherent,2 that is, sentential, infinitival complements that behave like finite embedded clauses with respect to extraction. Like Italian, they also have so-called coherent or restructuring infinitival complements that are transparent for several types of extraction processes. For instance, the direct object of the embedded verb can be moved into the matrix clause in restructuring infinitives (20a) but not in sentential infinitives (20b).



weil die Maria der Hans [t zu besuchen] versprach



since the Maria-ACC the Hans-NOM to visit promised



‘since Hans promised to visit Maria’



*weil die Maria der Hans den Peter [t zu besuchen] bat



since the Maria-ACC the Hans-NOM the Peter-ACC to visit asked



‘since Hans asked Peter to visit Maria’

There are two types of explanations in the literature for the transparency of restructuring infinitives. Many scholars (cf. Haider [1993], Napoli [1981], Rochette [1988], Rosen [1989, 1990], Rutten [1991]) have assumed that restructuring infinitives are monoclausal structures. In this account, the problem of transparency disappears trivially, though it reappears in the form of the question of under which circumstances two main verbs can project a single clause (see Cinque [2001] and Wurmbrand [2001] for an interesting new answer to this question). In biclausal accounts, their transparency (p.15) has been traditionally explained with the assumption that either restructuring infinitives are not full CP-complements or Verb Raising (VR) opens the domain for long-distance scrambling (LDS).

In this section, I want to discuss only the most crucial properties that distinguish coherent from incoherent infinitives and reserve more language-specific properties of coherent infinitives for discussion in later sections that are dedicated to an in-depth analysis of these constructions in the individual languages. Here I want to outline the core properties of coherent infinitives, which will provide us with a first circumscription of the problem of restructuring.

The most salient properties of coherent infinitival constructions are the following: First, coherent infinitives, as opposed to other (sentential and non-sentential) arguments, show a very restricted distribution in the sentence. Second, as already illustrated, coherent infinitives are transparent for several types of extraction processes. Third, the formation of a coherent infinitival construction is typically, though not necessarily, associated with a morphological effect that has become known as IPP (infinitivus pro participio)-effect.

1.3.1 The immobility of coherent infinitives

I will first discuss the restricted distribution of coherent infinitives. Coherent infinitives, as opposed to incoherent ones, cannot be extraposed. Subject-raising verbs like scheinen (‘seem’) obligatorily require a coherent infinitival complement, the extraposition of which results in ungrammaticality (21a).



*dass Hans schien [sich zu rasieren]



that Hans seemed himself to shave



dass Hans versprach [sich zu rasieren]



that Hans promised himself to shave

Subject control verbs like versprechen (‘promise’) may take a coherent or incoherent infinitival complement. Extraposition of the incoherent infinitival complement is fine (21b). The fact that incoherent infinitives may extrapose has been taken to show that these infinitives are full CPs, since it is assumed that only full-fledged sentential projections can be extraposed. Along the same lines, the fact that coherent infinitives cannot be extraposed has been taken as evidence that these infinitivals are smaller than CPs. They are often analyzed as IPs or sometimes even as VPs (cf. Broekhuis et al. [1995]).

At this point it is important to note that incoherent infinitives are not restricted to the extraposed position. Like other arguments, they can be scrambled into the top of the middle field. Example (22) shows that they may occur in a relatively high position in the middle field. In (22a), the infinitival has been scrambled across the subject. In (22b), the infinitival has been scrambled to a position just below the subject. However, sentences in which the infinitive occupies a very low position in the middle field are considerably degraded (22c).




weil [sie morgen zu besuchen] Hans gestern noch nicht versprechen wollte



since her tomorrow to visit Hans yesterday yet not promise wanted



‘since Hans did not yet want to promise yesterday to visit her tomorrow’



weil Hans [sie morgen zu besuchen] gestern noch nicht verprechen wollte



since Hans her tomorrow to visit yesterday yet not promise wanted



??weil Hans gestern noch nicht [sie morgen zu besuchen] versprechen wollte



since Hans yesterday yet not her tomorrow to visit promise wanted



*weil [sich zu rasieren] Hans schien



since himself to shave Hans seemed

Coherent infinitives, as opposed to incoherent infinitives, cannot scramble to the front of the middle field. The raising verb scheinen only allows a coherent complement, which cannot be scrambled at all (22d).

1.3.2 The transparency of coherent infinitives

Let us now look at the second salient property of coherent infinitives: transparency. Coherent infinitives, as opposed to incoherent ones, allow for long-distance scrambling. The arguments of an embedded infinitive can be scrambled into the middle field of the matrix verb. For instance, the embedded object can be scrambled across the matrix subject (23a). It is important to note that this process of long-distance scrambling, though it occurs most naturally with weak pronouns and anaphors, is not restricted to them. Example (23b) shows that full DPs that are arguments of the embedded infinitive can be scrambled across the matrix subject. Long-distance scrambling out of incoherent infinitives results in ungrammaticality. In (23c), the embedded object phrase das Buch has been scrambled across the matrix subject out of an extraposed, hence incoherent, infinitival complement that leads to ungrammaticality.



dass siei der Mann [ti zu besuchen] versprach



that her the man-NOM to visit promised



‘that the man promised to visit her’



dass [der Maria]i [das Buch]j Hans gestern [ti tj zu geben] versprach



that Maria-DAT the book-ACC Hans yesterday to give promised



‘that Hans promised yesterday to give the book to Mary’



*dass uns [das Buch] i Hans gestern bat [der Maria ti zu geben]



that us the book Hans yesterday asked to Mary to give



‘that Hans asked us yesterday to give the book to Mary’

As in English, the scope of non-wh-quantifiers and operators is restricted by clausal boundaries in German. Consequently, the sentences in (24) and (25), where the matrix verb bedauern (‘regret’) selects only an incoherent infinitival complement, may only have the narrow scope reading in (b) but not the readings with a (p.17) wide scope of the quantifier/operator represented in (c). Examples (24d) and (25d) illustrate again that incoherent infinitival clauses are also opaque with respect to (overt) extraction. The direct object, be it a personal pronoun like in (24d) or a negative quantifier as in (25d), may not be extracted from the infinitival and scrambled above the matrix subject.



weil er [sie nicht geheiratet zu haben] bedauerte



since he her not married to have resented



since he resented not having married her



since he did not resent having married her



??weil sie der Mann [ t geheiratet zu haben] bedauerte



since her the man married to have regretted



weil er [niemanden geheiratet zu haben] bedauerte



since he nobody married to have regretted



since he regretted not having married anybody



since for no x: he regretted having married x



??weil niemanden der Mann [t geheiratet zu haben] bedauerte



since nobody the man married to have regretted

However, a non-wh-quantifier/operator embedded in a coherent infinitive may take scope over the matrix clause. Thus, the sentences in (26) and (27), in which the matrix verb wagen (‘dare’) allows for a coherent infinitival complement, are ambiguous between the readings given in (b) and (c). In (26) and (27), I have somewhat impressionistically bracketed arguments that belong to the embedded verb within the infinitive, to highlight the effects of restructuring. In (26), these elements include the negative marker since it occurs between the infinitive and its direct object argument.



weil er [sie nicht zu küssen] wagte



since he her not to kiss dared



since he dared to not kiss her



since he did not dare to kiss her



weil er [ niemanden zu küssen] wagte



since he nobody to kiss dared



since he dared for no x: to kiss x



since for no x: he dared to kiss x

Finally, coherent infinitives, as opposed to incoherent infinitives, allow for “long extraposition.” I have shown in (14) that extraposition is clause-bound. In (14d), repeated here as (28a) for convenience, the matrix verb bestärken (‘encourage’) only (p.18) takes an incoherent infinitival complement headed by versuchen (‘try’), extraposition out of which leads to ungrammaticality. The verb versuchen, however, optionally takes a coherent infinitival complement. Hence, when it functions as matrix verb as in (28b), long extraposition becomes possible.



*dass der Lehrer [VP [VP die Kinder [CP t zu versuchen] bestärkt] [die Aufgaben zu lösen]]



dass der Lehrer [VP [VP die Kinder [? t zu bestärken] versucht] [die Aufgaben zu lösen]]

1.3.3 The Ipp-effect in coherent infinitives

The third salient property of coherent infinitives is the IPP-effect. As I said before, coherent infinitives typically, though not necessarily, display this morphological effect. For instance, in German, though not in Dutch, only bare infinitives, but not to-infinitives, give rise to the IPP-effect. The IPP-effect occurs when a verb that selects a coherent infinitive (the dependent infinitive) is put into a perfect tense (present perfect or past perfect tense). In this case, the verb does not show up in its expected past participial form but is realized as bare infinitive (the IPP-infinitive). Example (29) illustrates the IPP-effect in Dutch with the dependent infinitive being a to- infinitive. Example (30) illustrates the IPP-effect in High German, where the dependent infinitive must be a bare infinitive.



*dat Elsje hem een brief heeft gezien te schrijven



that E him a letter has tried-PART to write



dat Elsje hem een brief heeft zien te schrijven



that E him a letter has try-INF to write



*dass sie Hans nicht küssen gewollt hat



that her Hans not kiss want-PART has



‘that Hans has not wanted to kiss her’



dass sie Hans nicht hat küssen wollen



that her Hans not has kiss want-INF

I will discuss the IPP-effect in more detail in the different sections devoted to the individual West Germanic languages.

1.3.4 Conclusions

To summarize, coherent infinitives, compared to other (sentential) arguments, have a very restricted distribution: they can be neither scrambled nor extraposed; they are transparent for several types of extraction processes, including long-distance scrambling, quantifier/operator movement, and long extraposition; and finally, they typically give rise to the mysterious IPP-effect.

(p.19) 1.3.5 The formation of verb clusters in coherent infinitives

We have seen that while a coherent infinitive as a whole cannot be scrambled, parts of it, namely, its arguments, can undergo movement into the middle field of the selecting verb. It is important to note that the infinitival head itself cannot be separated from the selecting verb. The contrast in (31) shows that while the embedded direct object can be scrambled across the matrix subject, the infinitive itself cannot be scrambled across an adverb that modifies the matrix verb. In (31a–b), I have scrambled the direct object of the infinitive above the matrix subject to make sure that we are dealing here with a coherent construction. In this case, the adverb has to precede the infinitive and can modify both the matrix verb and the dependent infinitive, as can be expected from the general transparency of coherent infinitives (31b).



*weil sie der Hans zu besuchen oft versprach



since her the Hans to visit often promised



‘since Hans often promised to visit her’



weil sie der Hans oft zu besuchen versprach



since her the Hans often to visit promised



‘since Hans (often) promised to (often) visit her’

Finally, let us look at the behavior of coherent infinitives with respect to topicalization. First note that the head of the coherent infinitive and the verb selecting it can be topicalized as if they formed a constituent, while if the infinitive is incoherent its head cannot be topicalized with the verb selecting it. In (32a), versprechen (‘promise’), which optionally selects a coherent infinitival, can be topicalized with its dependent infinitive, whereas topicalizing the verb bestärken (‘encourage’) together with its dependent infinitive results in ungrammaticality, since this verb only selects an incoherent infinitival complement.



[zu besuchen versprochen] hat sie der Hans noch nie



to visit promised has her the Hans-NOM yet never



‘So far Hans has never promised to visit her’



*[zu besuchen bestärkt] hat mich Hans seine Schwester noch nie



to visit encouraged has me Hans-NOM his sister yet never



‘So far Hans has never encouraged me to visit his sister’

It should be noted, however, that although the selecting verb cannot be topicalized without its dependent infinitive in a coherent construction (33a), the dependent infinitive can be topicalized to the exclusion of its selecting head (33b). This contrast is reminiscent of the contrast that we observed in the behavior of particle verbs with respect to topicalization (cf. [13c–d] earlier): In a restricted way, it is possible to topicalize the particle without the verb it seems to be attached to, while it is impossible to topicalize the particle verb without its particle. Furthermore, it is important to note that topicalization of the dependent infinitive alone voids the IPP-effect, as is illustrated in (33c). (p.20) Voiding of the IPP-effect pertains only to perception verbs in German but occurs quite generally in Dutch and West Flemish (see chapter 7 for details).



*[sehen] wird sie Hans nicht kommen



see will Hans her not come



‘Hans will not see her coming’



[kommen] wird sie Hans nicht sehen



come will her Hans not see



[kommen] hat sie Hans nicht *sehen / gesehen



come has her Hans not see-INF / see-PART



‘Hans has not seen her coming’

The fact that the head of a coherent infinitive cannot be separated from its selecting verb (see [31a–b]), as well as the fact that the head of a coherent infinitival can be topicalized with the verb that selects it to the exclusion of the arguments of the infinitive have been taken as direct evidence for the assumption that coherent infinitival constructions involve the formation of a verbal complex by Verb Raising (VR). VR is a process of head movement that adjoins a dependent infinitive to its selecting verb (Evers 1975).

The formation of a complex head by either VR or reanalysis (Haegeman and van Riemsdijk 1986) has been taken as a causal explanation for the restrictive distribution and the transparency of coherent infinitives. The former property of coherent infinitives would follow trivially from the fact that the dependent infinitive and its selecting verb form a complex head. The latter property of coherent infinitives has been explained in the following way: Evers assumed that VR gives rise to a process of S-pruning. Van Riemsdijk and Haegeman assumed that reanalysis gives rise to a multidimensional representation: one dimension to represent the biclausal properties and one dimension to represent the monoclausal properties of coherent infinitives. In more modern terminology but still exploiting the same idea, we may analyze VR as a process that incorporates the dependent infinitive into the selecting verb. We may then assume with Baker (1988) that once the head of a projection has been incorporated, the entire projection, here the coherent infinitival complement, becomes transparent with respect to the incorporator.

VR has also been taken to be responsible for the IPP-effect. If it should turn out to be correct that VR actually is the trigger for the IPP-effect, then the data we have looked at so far suggest that we are dealing here with two independent factors for the following reason: We have seen that to-infinitives in German may be transparent for several types of extraction processes, but I have also pointed out that they never give rise to an IPP-effect. It thus seems that VR and the transparency of coherent constructions are independent factors and are not causally related to each other as has been assumed traditionally; VR may be a side effect of the transparency of coherent infinitives but not its trigger.

Also, my initial presentation of the core data shows that the evidence for the formation of a complex head is not as direct or clear-cut as it seems at first sight. The (p.21) fact that the head of a coherent infinitival cannot be separated from its selecting verb cannot only be explained by adjunction or incorporation but may simply follow from the inability of the head of a coherent infinitive to scramble. I have pointed out that constituents that are interpreted as predicates generally resist scrambling. Of course, this alternative account presupposes an explanation of why and how coherent infinitives are to be interpreted as part of the main predicate.

Furthermore, the argument that the fact that the dependent infinitive and its selecting verb can be topicalized together shows that these two heads form a constituent (32a) is considerably weakened by the observation that the dependent infinitive can be also topicalized without its selecting head (33b).

To summarize, the properties discussed earlier have been standardly interpreted as indicating that restructuring infinitives involve (1) verbal complexes and (2) a monosentential middle field. Monoclausal accounts assume that verbal complexes can be base generated. Haider (1993) assumes that verbal complexes are base generated head-adjunction structures. Wurmbrand (2001) argues, following Chomsky (1995b) in analyzing transitive verb phrases as involving two verbs—the verbal root and an abstract causative verb (small v)—that the verbs in a coherent construction share a single small v. Cinque (2001) argues that restructuring verbs in Italian are those verbs, in essence modal verbs and aspectuals, that can be assumed to be base generated in designated functional positions. All these accounts share the assumption that restructuring infinitival constructions involve only one VP, from which assumption it follows that restructuring infinitives have a monoclausal middle field, that is, have only one IP.

Biclausal approaches hold that each verb in a coherent construction projects its own clause (full or reduced) and assume special restructuring processes (S-pruning, reanalysis, Verb Raising) to account for the presumed monoclausal properties of coherent constructions.

1.4 Remnant movement

As already noted, the term remnant movement was coined by Den Besten and Webel-huth (1987) to account for a peculiarity of verb-preposing in German and Dutch. Provided that only XPs can move into XP-positions, it follows that what has been moved into [Spec,CP] in (34) is not simply a verb but must be minimally a full VP. Thus, they propose to analyze (34a) parallel to cases of VP-preposing (cf. [34b]) in which the direct object has been scrambled out prior to VP-to-CP movement. The moved VP is called a remnant category since it contains, at least, the trace of the direct object as is indicated in (34c).



gelesen hat Hans das Buch



read-PART has Hans the book



‘Hans has read the book’



[VP das Buch gelesen] hat Hans tVP



[VP tSCR gelesen] hat Hans [das Buch]SCR

(p.22) In the following subsection, I discuss the essential properties of and the relevant restrictions on remnant movement. The latter will be shown to follow from the workings of Attract Closest. Then, I discuss the interaction between remnant movement and head movement. I argue that remnant categories created by head movement cannot undergo further movement and show how this restriction can be derived from Attract Closest as well. Finally, I show that the original account of (34a) by Den Besten and Webelhuth is flawed by presenting data that show that the remnant VP in (34a) cannot be taken to have been created via scrambling. Instead, I argue that remnant VPs in German are created by licensing movement of VP-internal material into dedicated licensing positions in the lower middle field.

1.4.1 Properties of remnant movement

Remnant movement seems to be exempted from the Proper Binding Condition (PBC), which requires that traces be bound. It is typical of remnant movement to create unbound traces as we have seen earlier. In (34c), the fronted VP contains the trace of the direct object, which is not c-commanded by its antecedent.

Remnant movement typically also leads to so-called Anti-Freezing effects. A Freezing effect occurs if extraction of constituent a out of a constituent b takes place in a derived position of b. This is illustrated in (35a–b). In (35a), extraction of the wh-PP takes place from the base position of the direct object. In (35b), the direct object has been scrambled to a higher position from which extraction is excluded.



Worüberi hat keiner [ein Buch ti] gelesen



where-about has nobody a book read



*Worüberi hat [ein Buch ti] keiner gelesen



where-about has a book nobody read



‘Which topic has nobody read a book about’



[Ein Buch ti] hat darüberi keiner gelesen



a book has there-about nobody read



‘Nobody has read a book about that’

Example (35c) is a case of remnant movement. From a representational point of view, (35c) should be as bad as (35b), since the direct object in (35c) occurs in a derived position and contains a trace created by extraction. However, if we look at these cases from a derivational point of view, we immediately understand why (35b) is ungrammatical and (35c) is fine. The difference follows from the Extension Condition on derivations (Strict Cyclicity). In (35c), it is possible to extract the PP out of the DP in its base position and then move the remnant DP, obeying the Extension Condition, to a higher position. This derivation, however, is not available in (35b). If the PP were extracted first, movement of the remnant DP to a lower position would violate cyclicity. Thus, we see that obeying the Extension Condition voids—in a manner of speaking—Freezing effects but necessarily leads to unbound traces. Consequently, we may assume that the PBC is not a derivational constraint.

(p.23) Furthermore, as Müller (1996, 1998) has discussed at length in his important work on incomplete category fronting, cases of remnant movement display unexpected asymmetries. If we restrict ourselves to the interaction between remnant movement and scrambling, it is interesting to note that remnant categories can be topicalized while scrambling may not affect them, as is illustrated in (36). This asymmetry does not show up in cases of complete category fronting (37).



[zu lesen] hat das Buch keiner versucht



to read has the book nobody tried-PART



‘Nobody has tried to read the book’



?? dass [zu lesen] das Buch keiner versucht hat



that to read the book nobody tried has



‘that nobody has tried to read the book’



[das Buch zu lesen] hat keiner versucht



the book to read has nobody tried-PART



‘Nobody has tried to read the book’



dass [das Buch zu lesen] keiner versucht hat



that the book to read nobody tried has



‘that nobody has tried to read the book’

Müller (1996) concludes from these and other similar observations that remnant XPs cannot undergo Y-movement if the antecedent of the unbound trace has also undergone Y-movement, where Y-movement ranges over scrambling, wh-movement, and topicalization. In other words, a remnant category cannot, say, undergo scrambling if the operation that created the remnant category was itself a scrambling operation. Müller derives this constraint on remnant movement from his Principle of Unambiguous Domination, which is motivated by the need of traces, so Müller assumes, to be unambiguously identifiable.

There are several problems with Müller’s account though. First, there are cases where remnant categories can undergo scrambling, as we will see in chapter 4 (section 4.3.4). Second, various cases where remnant categories cannot scramble can be reduced to independent restrictions on the individual operations involved, as I will also argue in chapter 4. Third, Müller is assuming (following Den Besten and Webelhuth [1987]) that the remnant XPs in (36) are created by scrambling. In section 1.4.3, I will argue that remnant VPs are created by licensing movement of VP-internal arguments into dedicated positions in the middle field, which is to be distinguished from scrambling. Fourth, Müller’s Principle of Unambiguous Domination lacks conceptual plausibility in a theory of movement in terms of copy and delete, that is, in a theory where there are no traces to be identified.

From the point of view of feature checking, Müller’s observations indicate that for remnant movement to be possible two sets of features must be involved that cannot be checked in the same (type of) position. Thus, Müller’s generalization (to the extent that it is correct) can be derived from Attract Closest (Chomsky 1995b), as is (p.24) illustrated in (38). In order for a (remnant) category A to undergo scrambling, extraction of B out of A must involve a type of movement other than scrambling. In (38), the head F cannot select category B as a target since category A, containing and dominating B, is a closer target that shares the scrambling feature with B (cf. Fukui [1997]).3

  1. (38) FSCR … [A SCR … [B SCR …]]

There is a class of cases of remnant movement that are illicit although they obey Attract Closest (or Unambiguous Domination, for that matter). These involve topicali-zation of a remnant category out of which a category has been extracted via wh-movement as is illustrated in (39). Example (39a) is a case of topicalization of a clause across a wh-island, which leads to a mild, subjacency-like violation. Example (39b), where the wh-word is extracted from the embedded clause (creating a remnant category), however, is ungrammatical. The same contrast can be observed in English, as is shown in (40), taken from Pesetsky (2000). The relative grammaticality of (39c) is interesting in this respect. If the trace of the wh-word were contained in the fronted VP, as is standardly assumed, then (39c) should be on a par with (39b) and (40b). However, it is just as good as (39a). I will come back to this difference and show how it is to be explained in section 1.4.3.



??[dass Fritz Peter liebt] weiß ich nicht wer gesagt hat



that Fritz loves Peter know I not who has said



‘I don’t know who has said that Fritz loves Peter’



*[ dass Fritz t liebt] weiß ich nicht wen er gesagt hat



[that Fritz loves t] know I not who he has said



‘I don’t know who he said that Fritz loves’



??[geküsst] weiß ich nicht wen sie hat



[kissed] know I not whom she has



‘I don’t know whom she has kissed’



[give a book to John] I can guess who will



*[give a book to t] I can guess who Mary will



(I can guess who Mary will give a book to)

It is not clear why wh-movement differs in this respect from other types of movement. In other words, it is not clear why the PBC nevertheless seems to be relevant for wh-movement. Pesetzky (2000) assumes that there is a (special) command restriction on wh-movement.

Cecchetto (2001) argues that the illicit cases of remnant movement in (39b) and (40b) follow from the Phase Impenetrability Condition (cf. Chomsky [1998, 2001]). Contrary to Chomsky, Cecchetto proposes that the CP contains two escape hatches, one for wh-elements and one for D-linked elements like topics, and argues that elements that contain a wh-trace cannot make use of the escape hatch reserved for (p.25) D-linked elements. Thus, (39b) and (40b) involve movement across a strong phase in his system, while the topics in (39a) and (40a) can move via the extra escape hatch for D-linked elements. This seems like a promising approach to me, though it remains unclear how Cecchetto can account for the relative grammaticality of (39c). I will introduce the Phase Impenetrability Condition in chapter 6 and leave the question of the best treatment of the data in (39) open here, concluding from the preceding discussion that as long as we steer clear of extraction via wh-movement, remnant movement is freely available if Attract Closest and Strict Cyclicity are obeyed.

1.4.2 Remnant movement and head movement

In this section, I would like to address a further restriction on remnant movement that has been discussed in the literature (cf. Haider [1990], Sabel [1996.], Takano [2000]) and show how it can be reduced to the workings of Attract Closest as well.

So far, I have only discussed cases in which a remnant category is created by extraction via XP-movement. In the sentences in (41), the remnant category (topicalized in [41a–b] and clefted in [41c]) has been created by extraction via head movement.4



*[Ihr ein Buch t] gab Hans



her a book gave Hans



*[Het boek aan Marie t] gaf Jan waarschijnlijk



The book to Mary gave John probably



*It’s [a book to Mary] that John gave

All three cases involve remnant movement of the VP that contains a trace of the finite verb, which either has undergone V2 as in (41a–b) or has raised out of the lower VP-shell in a Larsonian structure in (41c). Takano (2000) argues that the data in (41) fall under the generalization in (42).

(42) Remnant movement of α is impossible if the head of α has moved out of α.

At this point the question arises what (42) can be derived from. It can be shown that the major restrictions on remnant movement follow from the mechanism of feature checking, more specifically from the restrictions on target selection. In (43), F and G are functional heads and XP is some category that contains Specifier A and Complement B. I make the standard assumption that XP and its head X share all relevant feature (i.e., XP is a projection of X). Furthermore, I assume that head and complement do not share features via agreement, while Specifier and head may share certain features via agreement.

  1. (43) G … F … [XP A [X B]]

If the functional head F attracts a feature of X (for checking), I assume that due to Attract Closest, movement of the entire XP will ensue (since the XP shares the feature with X (p.26) but is closer to F than X), unless the functional head F itself requires a head, that is, it is an affix.

If F attracts a feature of B and G a feature of X (not shared with B), we have a standard case of remnant movement. If X and B for any reason share the relevant feature with F, remnant movement is blocked, as has been discussed in section 1.4.1. If on the one hand, F attracts a feature of A that is not shared by X (a non-agreeing feature), then the Specifier will extract from XP. If on the other hand, F attracts a feature that A and X agree in, then movement of XP will ensue (a case of pied-piping).

Let us now look at the situation in which a remnant category is created by head movement. In this case F, being an affix, attracts a feature of X and X undergoes head movement to F, creating a remnant XP. The issue now is why the functional head G cannot attract the remnant XP. We have to distinguish two cases: (1) G attracts a feature shared by X. In this case, G will select F or FP but not XP due to Attract Closest. (2) G attracts a feature within the constituents of XP not shared by X. In this case, either A or B may extract from XP, but no movement of the remnant category XP will be triggered. It thus follows that movement of a remnant category that has been created by head movement is impossible.5

1.4.3 Remnant movement and licensing movement

As I have pointed out earlier, in the original account of Den Besten and Webelhuth (1987) of data like (1), the operation that evacuates the VP before it undergoes topicalization is assumed to be scrambling. English does not allow for the topicaliza-tion of remnant VPs. In their account this simply follows from the fact that English does not allow for scrambling. Ingenious though this account at first sight seems, it nevertheless turns out to be severely flawed at closer inspection. It suffers from two inconsistencies, the extraction paradox and the evacuation paradox that will be discussed and illustrated later.

To avoid these problems, I propose that the VP in German (and Dutch) is evacuated by licensing movement. In Hinterhölzl (1999), I argue that not only arguments but also VP-internal predicates move out of the VP to be licensed in specific positions in the middle field as is illustrated in (44).

  1. (44) [DPs [Neg0 [Focus0 [DPs [VP-adverbs [Pred0 [Asp0 [VPV ]]]]]]]]

In chapter 4, I will provide empirical arguments for the following licensing operations: Verb-particles are licensed in [Spec,AspP]. Small clause predicates, idiomatic expressions, and directional PPs are licensed in [Spec,PredP] above Asp0. Nominal arguments of the verb undergo Case-licensing movement to positions above VP-adverbs. From there they may undergo scrambling to higher positions according to their semantic/discourse properties. Thus, movement of arguments out of the VP is licensing movement that applies to all DPs independently of whether they are definite or indefinite and has to be distinguished from further movement that applies to DPs according to their semantic/discourse properties as is discussed in detail in chapter 2.

The assumption of licensing movement of VP-internal material that is to be distinguished from scrambling proper can solve the two long-standing problems with (p.27) the standard account of VP-topicalization in terms of scrambling and remnant movement, namely, the extraction paradox (1) and the evacuation paradox (2):

1. The assumption that there is licensing movement of VP-internal material to specific positions in the middle field is corroborated by the existence of so-called string vacuous scrambling as is illustrated in (45). In (45), the underlined phrases seemingly occur in their canonical position (in [45b] the direct object follows the subject) but have to be assumed to have been scrambled out of VP in order to be exempted from VP-topicalization.



[VP t gerechnet] hat wie immer keiner damit counted has as always no one there-with



‘as usual nobody has reckoned with that’



[VP t gelesen] hat gestern der Fritz ein Buch darüber read-PART has yesterday the Friz a book there-about



‘Fritz has read a book about this yesterday’

As (46) shows, these presumed scrambling positions do not behave like regular scrambling positions since they do not exhibit any Freezing effect (cf. [35] earlier) in that they allow for further extraction, which scrambles part of the phrase to a higher position. Thus, the data in (45) pose the following problem: If scrambling viewed as an operation that moves VP-internal material into the middle field were a unitary operation, then why is it that scrambled phrases allow for further extraction in certain positions but not in others?



gerechnet hat da wie immer keiner mit



counted has there as always noone with



gelesen hat darüber gestern der Fritz ein Buch



read has there-on yesterday the Fritz a book

2. Another problem for the standard approach is the fact that elements that resist scrambling can be left behind by VP-topicalization. This is illustrated for small clause predicates and indefinite “w-words” in (47a) and (47b), respectively. I will show in chapter 2 that scrambling proper may not affect (small clause) predicates. Furthermore, I will also demonstrate that indefinite w-words may only scramble to extend their scope domain. This is not the case in (47b). If scrambling were the only operation that can evacuate material from the VP, then it is hard to understand how these elements can be left behind by VP-topicalization.



gegessen hat der Karl das Fleisch roh



eaten has the Karl the meat raw



‘Karl has eaten the meat raw’



gelesen hat die Maria erst gestern was



read-PART has the Maria only yesterday something



‘It’s only yesterday that Mary read something’

(p.28) A possible solution to these problems is the assumption of licensing movement (prior to scrambling) that moves VP-internal material, irrespective of its quantificational or referential potential, out of the VP. This assumption immediately solves the evacuation problem—elements that do not scramble nevertheless leave the VP to be licensed in specific positions in the lower middle field—but also opens up the way toward a solution to the extraction problem. Given the distinction between licensing movement and scrambling, we may assume that licensing movement out of the VP does not give rise to a Freezing effect and that this Freezing effect is connected with the “Specificity effect” of the semantically motivated scrambling operation. We know independently that extraction out of specific DPs is illicit. Since DPs that have scrambled across sentential adverbs or negation (crucially not ones that have scrambled across VP-adverbs) exhibit a Specificity effect, the Freezing effect associated with these scrambling operations follows without further ado.

Finally, note that the trace within the VP in (39c) earlier, repeated here as (48a), does not behave like a Case-marked wh-trace; otherwise we would expect (39c) to be on a par with (39b) and (40b). However, (39c) gives rise only to a mild, subjacency-like violation and is on a par with (39a), as we have seen in section 1.4.1 earlier. This fact, too, follows straightforwardly, if we assume that there is licensing movement of DPs (and of VP-internal predicates) out of the VP that leaves A-movement traces in the VP such that the wh-word in the embedded clause can bind a wh-trace outside of the fronted VP within its clause, as is indicated in (48a). The analysis of (39c) in (48a) makes this case of remnant movement parallel to cases of movement of remnant categories that are created by movement of the subject to [Spec,TP] or another licensing position for subject, as is illustrated in (48b). Example (48a) thus only differs from (48b) in that it involves extraction of the topicalized phrase out of a weak whisland, which accounts for its slightly marked grammatical status.



??[ ti geküsst]j weiss ich nicht [weni sie hat ti [VP tj]]



kissed I don’t know whom she has



‘I don’t know whom she has kissed’



[fired t by the company] John indeed was

To summarize, I have argued that the assumption of movement of DPs and VP-internal predicates to specific licensing positions in the middle field, as illustrated in (44) earlier, can provide a solution to the extraction problem as well as the evacuation problem and gives us an explanation for the otherwise rather mysterious contrast between (39b) and (39c).

1.5 Theoretical assumptions

In the following, I will endorse the Minimalist Program (MP) in its programmatic sense. That is to say that the discussion and treatment of scrambling, remnant movement, and restructuring is cast in the spirit of MP, without endorsing all its tenets and principles to the letter. For instance, I do make use of agreement projections, though (p.29) they are argued to be non-minimalistic in Chomsky (1995b). However, I do not make use of multiple specifiers though they are part of minimalist accounts of Case-licensing and other phenomena and abstract away completely from assumptions made within the bare phrase structure framework (Chomsky 1995a). An important minimalist notion that I consider and adopt in the description and treatment of scrambling, remnant movement, and restructuring is the concept of the phase. In particular, I propose that employing the notion of phases may be useful in approaching the problems posed by extraposition and VP-topicalization.

Consequently, I will now outline the core ideas and concepts that I will adopt and later indicate in the account of individual phenomena where I differ from standard treatment. In my view, the basic weakness and strength of MP lies in its programmatic character. I think MP cannot be simply adopted since it is not a fleshed-out theory (by now) as, say, Government and Binding theory was. But I would like to endorse it in the way in which it approaches grammatical phenomena.

The core idea lies in the way in which it conceptualizes grammatical representations and their well-formedness. The levels of representations are reduced to the two interface levels, PF and LF. As a consequence, every principle has to be formulated in a way that it can be reasonably taken to apply at one of the interfaces (as an output condition) or in such a general way that it can be taken to apply at every step in the derivation.

Another important concept of the MP is that the grammaticality of a derivation may depend on the properties of another derivation. In particular, the MP invokes economy principles that compare derivations that involve the same lexical resources and select the most economical derivation.

In GB-theory, operations such as move alpha apply freely and a single derivation that results from the free application of such operations is checked against the constraints formulated in the different modules. In the MP, individual constituents have particular needs that motivate them to move and failure to meet these needs can lead to a derivation that is uninterpretable at one of the interfaces. In short, movement operations of constituents are motivated or triggered by the particular needs that they have, technically implemented by the types of features that they possess. This means, in turn, that every movement operation that is proposed in the account of a certain phenomenon needs to be justified by specifying the features, in the ideal case, the interface features, that motivates it. I will come back to this point, shortly. I consider this an important tenet of the MP, because it will prompt us to search for genuine explanations rather than for merely technical accounts of syntactic phenomena, insofar as we strive to provide derivations that are preferably triggered by (interpretable) interface properties.

To be concrete and to specify the needs that a particular constituent may have, a DP argument holds a thematic relation to a predicate, has a certain grammatical function in the clause, has scope over other elements in the sentence, may be interpreted as a variable bound by a particular operator within a certain domain, may be the topic or the focus of the sentence, and so on, possibly all or several of these at the same time. These relations are encoded by various means in the languages of the world, among which we find morphological marking, prosodic phrasing, word order, and intonational marking.

(p.30) It has been a leading idea in generative theory from the beginning that these different types of marking are incorporated into the syntax or dependent on syntactic structure. In minimalist terminology this means that these relations are either part of the computational system itself or the result after a complex computation at one of the interfaces. A sentence can thus be viewed as the most economic derivation that connects the individual partial representations that encode the different relations among the lexical items that have been chosen to express a piece of meaning, say, a proposition. A core ingredient in this system is the copy theory of movement. Chains formed by movement in the MP consist of a sequence of copies of the moved constituent. Where the chain is pronounced and where it is interpreted is decided by independent principles. This analysis of chains as a sequence of copies allows for reconstruction without any special operation (of putting lexical material back). In this manner, the copy of a DP argument within the VP may be said to designate its thematic relation, the copy within IP its grammatical function, and its copy within the CP-layer its discourse status in the sentence.

Locality principles that constrain movement seem to follow from a least effort strategy. This is captured in various economy conditions, most important among which is Shortest Move. The basic idea is that a constituent must not move beyond the first position of the relevant type, an idea that was first captured, though in different terms, as Relativized Minimality by Rizzi (1990). Another important economy principle that I will make use of is the principle of last resort. Last resort allows a certain operation only if all other options are prohibited. Do-insertion in English questions and negative clauses is an example of a last-resort operation that may apply to save the derivation in case the verb itself cannot move to C. To summarize, here are the most important tenets of the MP that I endorse in this book:

  1. 1. Movement is triggered and needs justification.

  2. 2. Grammaticality depends on a comparison of derivations.

  3. 3. Constraints apply only at the interfaces or throughout (uniformly at) the derivation.

  4. 4. Chains are analyzed as a sequence of copies, whose interpretation and pronunciation are determined by independent principles.

  5. 5. Derivations are constrained by economy conditions and phase based.

1.5.1 Narrow syntax and the role of Case

One idea or concept that I will not adopt is the assumption that the computation is driven by uninterpretable formal features predominantly or exclusively. Dealing with phenomena like scrambling and restructuring, I reject the approach that has become known as narrow syntax. In chapter 2, I will provide a comprehensive discussion of the empirical nature of scrambling phenomena in German and argue that scrambling is triggered (partially) by pragmatic features like familiarity/specificity.

Another issue with which I disagree is the role of Case in the MP. There are actually two assumptions connected with Case that I think are mistaken: (1) The assumption that Case is not interpretable at the interfaces, in particular, that it is uninterpretable at LF. In investigating scrambling phenomena, I will argue that Case is essential for (p.31) permuting arguments by marking their grammatical functions (see chapter 2). (2) The assumption that Case is invisible for the computational system. To evaluate this assumption, let us look at the role that Case plays in Icelandic and German.

We have seen earlier that basic word order in German is dependent on the particular verb class. One such class is constituted by the so-called Dative-Nominative verbs, which are stative psych-verbs that assign the experiencer role to the Dative argument and the theme role to the Nominative argument. The experiencer is hierarchically higher than the theme argument and, consequently, the unmarked word order is DAT > NOM in German, as is illustrated in (49a).



weil dem Hans das Essen schmeckt



since Hans-DAT the food-NOM tastes



Þer hefur likað maturinn



you-DAT have liked food-the-NOM



[T  [DP-DAT  [DP-NOM]]]

The same class is found in Icelandic, where the two arguments are realized in a parallel fashion as Dative experiencer and Nominative theme (cf. [49b]). The difference between Icelandic and German, however, is the different grammatical function of the arguments. In German, as is expected, the Nominative argument, though the lower argument, is the subject of the clause. Icelandic is special in that these Dative experiencers function as the subject of the sentence, as is shown by their binding properties and their behavior in control structures (cf. Zaenen, Maling, and Thrainsson [1985]).

How can we account for the difference in syntactic behavior between Dative experiencers in German and Icelandic? If we identify the subject as the argument that moves to or is attracted by the T-head, then the Icelandic case is straightforward: T will simply attract the closest element with a D-feature, namely, the higher argument marked with Dative Case (cf. [49c]). In German, the case is more difficult since T has to attract the hierarchically lower argument, to render the Nominative argument subject of the clause. I want to argue that one way to achieve this is to assume that T in German does not simply attract the closest element with a D-feature, but that the T-head can refer to the Case properties of a DP and will attract the closest argument marked with Nominative Case. This account of the difference between German and Icelandic implies that Case is visible in the computation in German.

This difference between German and Icelandic is very instructive since it shows that the visibility of Case is not simply a matter of the presence (German) or absence (English) of morphological Case. Thus, contrary to narrow minimalist assumptions, my representations will include Case-positions and other positions that relate to the semantic contribution and the pragmatic role of the constituents in the clause.

1.5.2 The universal base hypothesis and triggers for movement

In the course of this study, I will adopt the UBH (Kayne 1994). Applying the UBH to a language like German, as I have discussed and illustrated in section 1.2, requires (p.32) additional movement operations to derive the surface forms that in the traditional head-final account are taken to be base generated. To derive OV-surface orders from VO-base structures, I will provide additional (see section 1.4.3) empirical evidence that VP-internal predicates and arguments move out of the VP (to appropriate licensing positions in the middle field). To derive the monoclausal properties and verbal complexes of restructuring infinitives, I will make substantial use of remnant movement of diverse categories that include VPs and TPs. It is an important goal of this study to not only describe the movements necessary in a Kaynean analysis to derive the surface properties of restructuring infinitives but to also provide motivation for the complex derivations proposed at each step. Hence I will try to justify each movement operation by specifying its syntactic or interface-related triggers. Only to the extent that these triggers seem non-arbitrary and non-stipulative can the endeavor be considered successful that consists of combining the restrictive theory of phrase structure implied by the UBH with the restrictive theory of movement inspired by the MP.

At several points of this study, I will evaluate the results of this endeavor by comparing the new analysis with the traditional account, discussing their strengths and weaknesses. In my opinion, this endeavor is successful overall. However, some recalcitrant problems remain, foremost the issue of extraposition and the topicalization of verbal projections (extended verb phrases). With these issues the goal of this study has to remain more modest: In chapter 7, I discuss these issues, pinning down the problems they pose for this approach, and outline paths toward their solution. The solution I provide for VP-topicalization is very promising, while I have to leave a full account of extraposition for further research.


(1.) In (14), I assume that adjunction is to the VP. Whether this is correct or whether adjunction has to apply to the local IP is immaterial to my purposes here.

(2.) The term kohärenter Infinitiv (‘coherent infinitive’) was introduced by Bech (1955), who, to my knowledge, was the first to describe in a fairly sophisticated and formal way the properties of this type of infinitivals. To honor Bech’s pioneering work, the term coherent is standardly used within the German linguistic literature on the subject. Within the Romance tradition of work on the subject (cf. Burzio [1986], Rizzi [1978, 1982]) restructuring infinitives became the standard term of reference. I will use both terms interchangeably in this book.

(3.) An anonymous reviewer points out that the relevant restriction on remnant movement, namely that the remnant creating and the remnant moving operations cannot be of the same type, falls out as a special case of a general restriction derivable from Williams’s (2003) recent work on representation theory and that Attract Closest seems insufficient to characterize all cases of illicit remnant movement not filtered out by strict cyclicity. The reviewer is correct in this position, and I show in chapter 4 that certain cases of illicit remnant scrambling are independently ruled out as cases of illicit scrambling of predicates. The point of (38) is to show that Müller’s restrictions on remnant movement can be interpreted as cases of an A-over-A effect, which can be cast in the MP as a violation of Attract Closest. The reviewer is also correct in his point that the application of Attract Closest must be limited in a way as to allow, for instance, for the parallel movement of subject and object out of the VP into licensing positions in the IP. This issue will be dealt with in chapter 4.

(4.) Note, however, that Haegeman (2003) claims that parallel cases are marginally possible in West Flemish (cf. [i]).

(5.) Pearson (2000) assumes that if the head of an XP is extracted, even non-agreeing features of the Specifier of XP may percolate up to XP. Under these assumptions, remnant movement of the entire XP is possible again. Pearson argues that the trace of the head is featureless and may therefore not give rise to the projection of features up to the maximum XP. While this seems plausible in an approach where traces are left behind by movement, it seems counterintuitive in a copy theory of movement, where the antecedent and its “trace” are copies of each other. Note that if such an approach is adopted as in Haegeman’s (2001, 2002b) account of the SOV order in embedded clauses in German and Dutch, another explanation has to be given for the data in (41).