Beginning with Chomsky 2001, numerous attempts have been made in recent years to offer phase-based analyses of the properties of Germanic Object Shift (OS), in particular Holmberg's Generaliz-ation (HG): the inability of the object to cross the finite verb (or, arguably, any phonologically visible category other than adjuncts, as elaborated in Holmberg 1999; see below). Some of these approaches have sought to capture HG as a PF effect due to cyclic linearization at the v/VP phase level (that is, HG as order preservation; see Richards 2004, Fox & Pesetsky 2005). In this well written and carefully argued monograph, Gema Chocano adheres strictly to Chomsky's (2001) model of the phonological conditioning of HG by cyclic spell-out, strengthening the validity of this model from the perspective of a related phenomenon in OV Germanic, namely scrambling in German. She uncovers subtle empirical evidence that the two central components of the Derivation By Phase analysis of Scandinavian OS—the phonological dislocation rule Disl, and the sensitivity of OS to the notion of phonological border—are at work in German scrambling too, thus allowing a unified treatment of OS and scrambling across VO and OV Germanic to emerge. Furthermore, the resulting treatment of Germanic OS in terms of these phonological properties is argued to solve a number of long-standing empirical issues left open by previous syntactic approaches to Germanic scrambling. I review some of Chocano's major claims below, pointing out that, while largely convincing, there are nevertheless some concep-tual problems with the chosen analysis that go unchallenged and unmentioned in this work.

The book comprises six chapters, of which the first four are exclusively devoted to an extensive empirical overview and a comprehensive review of the vast existing literature on the subject of German scrambling. Chapter 1 contrasts German(ic) scrambling with other kinds of scrambling found in other languages, concluding that OS in North Germanic and scrambling in West Germanic are essentially the same phenomenon, despite the fact that HG appears to constrain only the former.Footnote ¹Chapter 2 sets out some basic assumptions about German syntax that are relevant to the subsequent discussion, while chapter 3 reviews those syntactic, semantic, and phonological properties of German scrambling that an adequate analysis must be able to account for (such as whether or not scrambled constituents constitute islands for further extraction; the unstressed (or destressed) nature of scrambled con-stituents; and the consequent presuppositional (defocused) interpretation of scrambled constituents). Chapter 4 summarizes the major previous accounts of German scrambling, arguing that none of the existing syntactic approaches—whether they treat scrambling as movement (A or A-bar, substitution, or adjunction) or as base generation—are able to account for the full range of data reviewed in the previous chapter.

Ultimately, three main sets of data are identified as being particularly problematic for both movement and base-generation accounts, and these form the empirical challenge that the analysis presented in chapter 5 is designed to meet. Let us therefore review chapter 5 by looking at each of these briefly in turn.

The first set of data involves VP-topicalization:

1. (1)

   1. a.

   2. b.

   3. c.

The contrast between 1a,b and 1c is argued to undermine purely base-generated accounts of scrambling, such as Fanselow 2001 and Haider & Rosengren 2003, since all the fronted VPs in 1 should be base-generable orders and thus all equally frontable. This suggests that diese Zigarren in 1c has moved above the adverbial to a vP-external position, outside of Diesing's (1992) domain of existential closure (assuming, following Wurmbrand 2001, that TPs cannot be fronted or topicalized). Further-more, 1a,b indicate that v/VP-internal scrambling must be possible in German, with the existential subject remaining in situ in 1a and the accu-sative DP scrambling VP-internally across the dative in 1b; this is confirmed by the acceptability of 2a,b (from Fanselow 2001).

1. (2)

   1. a.

   2. b.

Assuming the existential subject to remain vP-internal here (that is, inside Diesing's domain of existential closure), both the ACC>DAT and the DAT>ACC orders must be due to VP-internal processes. To account for these and related facts, Chocano's first proposal in the book (finally arriving on page 223) is that there are two kinds of clause-internal, non-focus scrambling in German: one is the result of different base-generation options for arguments within VP, as diagnosed by the possibility of post-subject scrambling in 2, and the other is movement out of vP, as diagnosed by the impossibility of fronting such a scrambled constituent along with v/VP, as in 1c.

The VP-internal nature of the discourse-given DPs in 2 relates to the second data set that Chocano identifies as problematic for existing accounts.

1. (3)

   1. a.

   2. a’.

   3. b.

   4. b’.

Whereas the destressed, discourse-given accusative DP in monotransitive predicates must undergo scrambling, as in 3a,a’, it may remain in situ in ditransitive predicates (thus 3b,b’ are fine). This poses a problem for the interface-economy approach to base generation of Neeleman & Reinhart (1998), in which 3b should always block 3b’ as the less costly option: 3b does not involve destressing of the discourse-given DP das Buch in the nuclear stress position (das Buch being the most deeply embedded constituent on the recursive side in 3b’) with concomitant relocation of stress onto dem Kind.

It is here, Chocano argues, that we have evidence for the relevance of Chomsky's (2001) phonological border in German scrambling (and thus, by extension, for the generalized phonological HG of Holmberg 1999). The phonological border of an XP is, essentially, the highest position containing overt (that is, phonologically visible) material within that XP. Chomsky's (2001:35) analysis of OS and HG is then based on the following assumptions:

1. (4)

   1. a.

   2. b.

   3. c.

It follows from 4 that movement to the vP phase edge is licensed (via EPP-insertion) wherever VP-internal arguments incompatible with Int’, such as weak pronominal objects or specific, discourse-given DPs, occupy the phonological border of v*P, as obtains when the verb is finite and moves out of VP (e.g. to CP in verb-second clauses); that is, HG is derived. If Chocano is correct, then this same effect can be observed in 3: in ditransitive 3b’, the accusative DP is not at the phonological border of vP owing to the presence inside vP of the higher, dative DP. Therefore, Int’ is not imposed on the accusative DP here (both Int and Int’ are compatible with the non-border position), and raising to the vP edge is not sanctioned by 4c. Thus, 3b’ is exactly analogous to the lack of OS in a Scandinvian non-HG environment; that is, those cases where the verb remains inside VP (such as when the verb is nonfinite) and a definite or pronominal object remains postverbal. The other option, exhibited by 3b, now provides further evidence for the availability of base-generated, VP-internal scrambling (as argued by Chocano on the basis of facts such as 1–2 above): here, the accusative DP das Buch is plausibly generated above the dative dem Kind inside v/VP and thus does occupy the phonological border of v*P; consequently, EPP-movement to the phase edge and across the adverbial is licensed, since the unshifted discourse-given das Buch would receive the incompatible semantic complex Int’.Footnote ²

The third data set that poses problems for existing accounts of scrambling instantiates what Chocano calls the “Freezing/Anti-Freezing paradox.” As is well known, and as extensively surveyed by Chocano in her review of the previous literature, there is contradictory evidence as to whether scrambled constituents in German constitute islands for sub-extraction, and thus for whether scrambling should be viewed as base generation or movement: Freezing effects suggest movement is at stake, whereas Anti-Freezing (transparency for extraction) plays into the hands of base generators. Setting aside the conflicting evidence from NP-PP splits, was-für splits and split topicalizations whose variable behavior regarding subextractability Chocano attributes to other factors (following Wiltschko 1997, de Kuthy & Meurers 2001, Fanselow & Ćavar 2002, and others), Chocano hones in on scrambled coherent infinitives as providing the greatest challenge for existing approaches. Whereas topicalization and wh-movement to spec-CP are possible out of scrambled infinitives, as in 5a, “double scrambling” (that is, scrambling out of the scrambled constituent) is only possible with pronominal arguments, as shown in 5b,c (this observation and the data in 5b,c are due to Müller 1998).

1. (5)

   1. a.

   2. b.

   3. c.

Unless pronominal scrambling is assumed to be a different kind of movement from other instances of DP-scrambling (as suggested by Müller 1998), something which would run counter to the main stated objective of treating all cases of OS and scrambling in German and Scandinavian as instances of a single, identical phenomenon, then purely syntactic accounts are unable to explain the contrast in 5b,c. By contrast, the facts in 5 are readily amenable to an account that recognizes the role that phonological features and processes play in OS/scrambling, such as Chomsky 2001. To this end, Chocano argues that the paradigm in 5 provides evidence for Disl, a central device in Chomsky's (2001) system but one which has been almost completely overlooked since its original postulation (aside from Svenonius 2001, who rejects it as ad hoc and unnecessary)—indeed, it is never mentioned again in any of Chomsky's subsequent works (Chomsky 2004, 2007, 2008). Disl, like Th/Ex in the same work, is claimed to be a phonological operation that strips away phonological features from certain syntactic objects within vP at points that occur prior to the next strong phase level (that is, prior to CP), moving those phonological features into a higher position. In the Derivation By Phase system, any language with the OS property (that is, 4a) must also have Disl on theory-internal grounds: the raised object in spec-vP should act as a defective intervener that blocks Agree between the head T and the subject DP in the lower spec-vP, rendering impossible the valuation of T's ϕ-probe (since the intervening object DP is inactive), Case on the subject, and arguably also the EPP-feature on T (since the subject cannot now raise to spec-TP):

1. (6)

On the further assumption that “Inactive trace disallows Match” (Chomsky 2001:28), Disl is invoked in order to rescue the OS structure in (6), allowing Agree(T, Subj) to take place and thus the structure to converge.

Disl would take place upon completion of the vP phase (we can think of it as a kind of early PF-Transfer of inactive material in the vP phase edge). If German scrambling is another instance of OS (that is, 4, yielding 6), then it is expected that Disl should apply in German scrambling structures too at this point. Chocano argues that the divergent behavior in 5a–c follows if German scrambling indeed involves Disl. She proposes that, as a phonological dislocation rule, Disl requires “full phonological integrity” (p. 251). Essentially, this means that Disl cannot apply to incomplete constituents, such as those containing a trace. The difference between 5a and 5b,c then emerges, since only in the case of double scrambling (5b,c) is the object that is contained inside the scrambled infinitival a trace (that is, its phonological features spelled out) at the point when Disl applies to the infinitival phrase, owing to the fact that the object itself undergoes separate scrambling, and thus its own application of Disl at the vP level. Further, as a phonological process, Disl is expected to interact with other components of the syntax-PF mapping procedure, such as prosodic phrasing (see Nespor & Vogel 1986, Selkirk 1995). Chocano claims that this is what is responsible for the difference between 5b and 5c: the weak, phonologically deficient pronoun es undergoes phonological restructuring to become part of the preceding phonological phrase comprising the scrambled infinitive, thus rendering the latter phonologically complete again.Footnote ³ Thus repaired, the phonological phrase {zu lesen es} is a viable candidate for Disl. In 5b, by contrast, no such restructuring with zu lesen is possible owing to das Buch's forming its own phonological phrase.

Aside from a few technical details relating to the timing of Transfer and the nature of Spell-Out, the analysis Chocano offers and the case she makes for maintaining Chomsky's Disl and phonological borders across Germanic are convincing and successful.Footnote ⁴ By adopting Chomsky's (2001) approach wholesale and demonstrating its validity for an OV Germanic language, her analysis inherits all the virtues of the Derivation By Phrase approach, including its blind, modular conception of OS/ scrambling as a narrow-syntactic operation (triggered by EPP-features, in line with 4b,c) that is not driven or caused by the semantic/pragmatic or phonological properties of the items that shift but rather has semantic/ pragmatic and phonological consequences at the interfaces. However, in its often uncritical adoption of that approach, Chocano's analysis also inherits its problems. I briefly mention some of these in the following paragraphs.

Firstly, both Disl and phonological borders are conceptually rather dubious devices. In the case of the latter, it is unclear why a semantic rule (the assignment of the interpretive complex Int’ at the semantic interface) should be sensitive to and conditioned by phonological features (which should be absent from the semantic interface). Further, as pointed out in Richards 2004:29, there is a contradictory requirement in the rules as stated in 4 above, in which the phonological border of vP is assigned Int’ whereas the edge of vP is assigned Int, since in the relevant cases (that is, where the object shifts/scrambles into the edge of vP), the edge becomes the phonological border (post-movement). Assuming movement to take place before Transfer to the semantic interface, this implies that the interface must have some way to tell a derived phonological border (that is, edge) from a pre-movement one if each is to be assigned complementary semantics: somehow, Int must presumably override Int’ just in the case where border = edge. It is also questionable whether the object can ever be said to be in the phonological border of vP at the point at which EPP-insertion, and thus OS/scrambling, should apply in the syntax, since the subject in spec-vP will always occupy the vP border prior to object raising, thus preventing 4 from ever applying to the object.

Turning next to Disl, its anomalies as a spell-out rule are readily apparent. Unlike all other cases of Transfer, which are regulated by the Phase Impenetrability Condition (PIC), spell-out by Disl uniquely targets the edge of the phase (spec-vP) rather than the complement domain of the phase head; it transfers the targeted constituent only to PF, not to the semantic interface as well; and it happens early, as soon as the lower phase (vP) is completed, rather than waiting until the next higher phase head is merged, that is CP, as under the Derivation By Phase version of the PIC (which regulates all other aspects of this system). Far from being a “fundamental tenet” (p. 211) of the minimalist framework, Disl would therefore seem to be a departure from the Strong Minimalist Thesis and the assumption of optimal design. The only conceptual argument in Disl's favor is its purported indispensability for rescuing the violation of the Minimal Link Condition in 6 above, allowing Agree to obtain between T and the subject in spec-vP. However, the problem in 6 only arises if defective intervention (DI; Chomsky 2000:123, 129) is a real phenomenon. As argued extensively in Richards 2004:161ff., defective intervention—that is, the notion that inactive nominals may intervene for Match/Agree—might itself be a departure from optimal design, since it undermines the Activity Condition on Agree and thus deprives us of a rationale for why Case features should exist in this system. In the absence of defective intervention, by far the simplest approach to 6 is to assume that the inactive (Case-valued) object in spec-vP simply does not intervene at all, owing to its no longer being active. Aside from the configuration in 6, Chomsky (2000, 2001) offers just two additional pieces of evidence that inactive nominals induce intervention effects, namely the ban on superraising (Chomsky 2000:128–129 (47)) and the lack of long-distance nominative agreement across in-situ quirky dative subjects in Icelandic (Chomsky 2000:130–131 (51)). Both of these cases have simpler and more plausible explanations that do without defective intervention, principally involving the PIC (see Richards 2004:161ff. for details). Given the tenuous conceptual and empirical support for defective intervention, then, it is far from a given that Disl is “absolutely necessary” (p. 211) in order to rescue the structure in 6. It would, in this light, be a redundant extension to the OS/scrambling mechanism in 4.

Finally, it is taken for granted that the empirical generalization on which the Chomsky 2001 analysis of HG and OS is based is correct; namely Holmberg's (1999) extended version of HG in which not only the verb but all other overt non-adjunct categories block OS (a “fundamental insight,”, p. 210). This version of HG has given rise to some highly insightful and influential analyses of OS, such as Fox & Pesetsky's (2005) attempt to derive HG as a cyclic PF-linearization effect imposed at the VP phase level (a work that Chocano overlooks). However, as argued extensively and, to my mind, completely convincingly by Bobaljik (2002) in his seminal PF-based, (morpho-) phonological account of OS (another important work that Chocano neglects to mention), the unified HG of Holmberg 1999 misses some important generalizations holding across OS and passives in Scandinavian. In particular, in those dialects in which OS of the direct object is possible across indirect objects and/or particles, so is passivization; where OS is impossible across these elements, so again is passivization. Furthermore, as Bobaljik (2005) points out in his reply to Fox & Pesetsky (2005), the very fact that there is such variation across Scandinavian dialects as to whether object movement is possible across indirect objects and particles is highly suggestive that these movements constitute a qualitatively different phenomenon from the HG effect that obtains between the verb and its direct object, which is absolute and invariable. For such reasons, Richards 2004 proposes that “HG proper” (that is, the ban on the object crossing the verb) is a PF-imposed linearization effect arising within the vP phase; as such, it reduces to an inviolable interface requirement (the inescapable need for structures to be linearized). By contrast, the variable restriction on arguments crossing other arguments, such as direct objects shifting or scrambling across indirect objects, is indicative of syntactic minimality effects (Attract Closest, Minimal Link Condition, etc.), which may be obviated in various ways: equidistance, base generation of different orderings, dis-placement of the intervener (see Anagnostopoulou 2003), tucking-in (N. Richards 1999), etc. In this connection, it is perhaps telling that Chocano herself has to appeal to such notions in order to explain why Dutch scrambling does not allow reversals of ordering between indirect and direct objects whereas German scrambling does (she invokes N. Richards's tucking-in to derive the former (pp. 276–277), and base generation to explain the latter (see above)). The status and relevance of the Holmberg's (1999) unified HG is thus questionable even in the present work, except for the semantic/pragmatic argument from focus and unmarked word order in ditransitives (see discussion of 3b’ above).

In sum, some critical evaluation of the adopted framework covering such issues as those suggested above would not have gone amiss. Nevertheless, despite these reservations, this book is an accomplished work, which can be recommended for its compelling and detailed attempt to substantiate Chomsky's (2001) OS analysis and thus the phonological HG on which it is based (Holmberg 1999). It offers an interesting and appealing elucidation and exploitation of two of the less regarded properties of the Derivation By Phase system—Disl and the phonological border—and argues persuasively for the applicability of these ideas to German and OV Germanic more generally. This is certainly an important, and welcome, result.