1. Introduction to the Derivations & Evaluations (D&E) Framework

Over the past thirty years or so, the field of generative syntactic theory has witnessed (and, in my opinion, benefited tremendously from) the birth of differing approaches to defining the structure of natural language syntax and the cognitive state of the human mind in relation to language. The most recent instantiation of the Principles & Parameters (P&P) framework, namely, the Minimalist Program (MP), has ushered in a stage in theoretical development with a heavy focus on the conceptual (and philosophical) nature of language. According to Chomsky (1995:5–6), this is a natural progression if one buys into the core tenets of the P&P framework:

Further developments in the MP have led to the conclusion that certain phenomena and structures are best accounted for by the mental facilities external to the syntax, namely, the sensori-motor (S-M) and conceptual-intentional (C-I) interfaces, commonly referred to as PF (phonological form) and LF (logical form). Recent work in the MP has given way to claims that, to some extent, transfer the responsibility of descriptive adequacy to the interfaces. Other theoretical approaches, such as Lexical-Functional Grammar (LFG), Head-Driven Phrase-Structure Grammar (HPSG), and Optimality Theory (OT) among a host of others, continue to make use of representational-level filters and constraints. Such filters and constraints are theoretical constructs that current mini-malism has sought to eliminate in order to shift the conceptual focus of the theory from the interfaces to the syntax. The ideal minimalist system should, then, along these lines, further have the property that “no filters are imposed on the end products of derivations, and no global filters (e.g. comparison of derivations) assign status to derivations as a whole” (see Frampton & Gutmann 2002:90). To employ OT-terminology, in main-stream minimalist theory, the component of the grammar that generates candidates (GEN) is powerful enough to generate “perfect” derivations without the assistance of global filters. In this regard, an evaluator (EVAL) is commonly taken to consist of universal principles and language-specific filters (often regarded to be external to the syntax). As pointed out by Pesetsky (1997, 1998) and Dekkers (1999), the advantage of OT is that such principles and filters can be more adequately expressed by means of the ranking of more primitive violable constraints.

On the surface, then, it suffices to say that the MP and OT appear to be mutually exclusive and should be viewed as competing theories. In some respects, the contrast between the MP and OT and their respective adherents and detractors is remarkably similar to the ancient philosophical “lumpers” versus “splitters” dichotomy (going back to Aristotle and Plato). The lumper/splitter issue arises when there is the need to create classifications and assign examples to them. Accordingly, recent developments in the MP can largely be viewed as an attempt at lumping, that is, as an attempt to create the most restrictive theory possible by lumping individual phenomena within larger mechanisms and principles.

Naturally, in such a gestalt view, the finite differences that exist in languages and dialects are housed under the larger, more general mechanisms. In contrast, OT-supporters employ the splitting-technique that exploits the finite inter- and intra-language differences among different languages and language families through the interplay of (represen-tational) constraints and crucial, harmonic ranking of these afore-mentioned constraints.

Although the distance between lumpers (MP-adherents) and splitters (OT-adherents) is admittedly gradient, the predominant view remains that these two frameworks and ways of interpreting the faculty for human language (F_L) contrast rather than complement one another. The long-standing arguments put forward by each camp are well known. Lumpers assert that GEN can be constructed and managed in a way that does not require global filters to evaluate—or worse, manipulate—units created by GEN. According to this approach, the OT-view boils down to little more than “representation massaging” and fails to deliver universal claims about human language. On the other side of the coin, splitters point to natural data to demonstrate that even the best constructed GEN fails to account for all of the variation found in languages. Under such a view, a syntactic theory based solely on GEN comes at the expense of one that abandons descriptive adequacy on many levels.

In this monograph, Hans Broekhuis attempts to be a lumper and a splitter at the same time by combining core aspects of the MP and OT theory in his Derivations & Evaluations (D&E) framework. Broekhuis demonstrates that this idea and the intricate design of the D&E framework are quite innovative and novel for modern linguistic theory; yet the general idea of combining (GEN) and an evaluating component of computation of human language (C_HL) that employs violable and elastic filters (EVAL) goes back to Chomsky & Lasnik's (1977) Filters and Control. According to Chomsky & Lasnik (1977:44–45),

To solve the tension between these two poles, they propose that “there is a theory of core grammar with highly restricted options, limited expressive power, and a few parameters” next to a more peripheral system of “added properties of grammar,” which Chomsky & Lasnik “think of as the syntactic analogue of irregular verbs” (p. 47). Filters in such a framework are not viewed as a burden to the computational system. On the contrary, they are seen as contributing to the simplification of transformational rules:

In the D&E framework, derivations are created by making use of the operations (EXTERNAL) MERGE (binary union of two lexical items, LIs) —which Broekhuis refers to as SELECT in this book—and MOVE (or INTERNAL MERGE, that is, the displacement of an LI previously introduced to the derivation). The generation of syntactic hierarchical structure is thus taken care of by MERGE and MOVE, and critically, these particular operations are not violable (but the order in which they occur may be subject to variation).Footnote ¹ Another (somewhat) unique attribute of the D&E framework is its elimination of any sort of Numeration that serves as a feeder to derivational operations. Following previous work with Wim Klooster (Broekhuis & Klooster 2001), Broekhuis argues that the notion of a numeration should be eliminated. He adopts Grimshaw's (1997) claim that the candidates in the candidate set have the same meaning.

The clear-cut advantage of restricting GEN along these lines is that it solves a recurring dilemma in the OT-syntax literature, namely, that of defining a proper input and arriving at the input structure (see Vogel 2004). There are, of course, other constraints found in the narrow syntax, such as the economy constraints *MOVE, NOLEXM (do not move lexical verbs), NOVACM (do not apply INTERNAL MERGE when it is not visible in the output), and EPP(F) constraints that regulate the movement of LIs based on the featural identity of Probes and Goals. These constraints are accompanied by constraints situated at the external interfaces that act upon the entire representation, such as alignment and scope constraints. The interface conditions are, of course, much more varied in nature in the D&E framework as well as in standard OT.

The D&E framework, as championed here by Broekhuis (p. 56), suggests “in order to arrive at a descriptively and explanatorily adequate theory, restrictions must be placed on both the syntactic derivation and on the resulting syntactic representations.” Clearly, this is not the first time that a hybrid theory has been proposed uniting the EVAL component of OT with another formalism/framework. In a series of papers (1996, 2000, 2001), Bresnan develops a hybrid LFG-OT framework where the LFG component significantly restricts the candidate set created by GEN, thus producing a reduced—and in many respects, more manageable—candidate set to be evaluated by crucially ranked constraints.

A detailed comparison of Bresnan's LFG-OT and Broekhuis' D&E framework simply cannot be undertaken in a book review, but it cannot be denied that the purpose of incorporating minimalist machinery into the D&E framework is similar (if not identical) to the function LFG plays in hybrid LFG-OT approaches: to restrict GEN. Critically, it is not just any version of minimalism that is appropriate for the D&E framework; as noted by Broekhuis, with the introduction of Minimalist Inquiries (Chomsky 2000 and subsequent work), overt movement became totally superfluous from the perspective of the core grammar (due in large part to the adoption of Probe-Goal Agree relations for feature-checking). As a result, the derivation of LF- and PF-representations proceeds in fully parallel fashion.

Earlier stages of minimalism (ca. 1995) made use of global economy conditions (for instance, Fewest Steps and Last Resort) as well as feature strength and the overt versus covert movement dichotomy. Syntactic derivation proceeded in a nonparallel fashion with regard to its creation and subsequent evaluation of LF- and PF-representations. The adoption of the theory of strong phases and Agree since the earlier stages of the MP has forged a version of minimalism that employs violable interface constraints in a framework such as the D&E. In this monograph, Broekhuis takes on the task of providing a detailed account of Object Shift (OS) in the Germanic languages. In the following section, I shall introduce the nuts and bolts of Broekhuis' analysis of OS in the Germanic languages within the D&E framework.

2. Scope and Content

Chapter 2 develops two hypotheses concerning OS and scrambling. The first hypothesis contends that OS and scrambling are essentially the same type of movement, namely, A-movement. Accordingly, movement is triggered by the unvalued nominal features of the verb. The second hypothesis contends that there are in effect two subtypes of OS: The first type is A-movement triggered by the φ-features of the verbal root V, and the second type is A-movement triggered by the case features on the light verb v:

1. (1)

   

To combat claims in the literature that scrambling is a form of A'-movement or a mixture of both (see, for example, Webelhuth 1989, 1992), Broekhuis advances a tenable solution in assuming that scrambling is not a unitary phenomenon (as argued by Déprez 1989 and Mahajan 1990, among others), and that A'-movement that appears to resemble A-scrambling is nothing more than focus or neg-movement.

In this chapter, Broekhuis also wrestles with the traditional chestnut of determining whether West Germanic languages that license scrambling exhibit an OV or a VO base order. In the end, he sides with Kayne (1994) and Zwart (1997) that West Germanic exhibits an underlying VO-order. This assumption, namely, that all West Germanic languages are underlying VO-languages, forces Broekhuis to claim that all Germanic languages have obligatory short OS, that is, movement of the object triggered by the φ-features on the verbal root V. As a result of the assumption, Broekhuis proposes that the constraints below in 2 and their subsequent rankings in 3 (= Broekhuis' 108, p. 117) are responsible for determining the surface VO- and OV-order in Germanic languages.

1. (2)

   1. a. H-COMPL: Disfavors output representations in which the underlying order head-complement is inverted.

   2. b. EPP(φ): These constraints favor application of movements that satisfy the Last Resort Condition. If this constraint outranks the economy constraint *MOVE, we predict that the φ-features on the verbal root V trigger short object shift, and that the φ-features on T trigger movement of the subject.

   3. c. *STRAY FEATURE: This constraint requires, among other things, that the verbal features on the functional heads in the extended projections of the verb attach to the verb. If this constraint outranks the economy constraint *MOVE, we predict that the verbal features trigger V-to-v/Asp and v-to-I.

   4. d. *MOVE: Don't apply INTERNAL MERGE.

2. (3)

   

The crucial ranking of EPP(φ) >> *MOVE determines that all Germanic languages license short scrambling. The distinction between the VO- and OV-languages has been argued to follow from the fact that “in the former the verbal root V obligatorily moves into v/Asp, whereas in the latter it may stay in its base position” (p. 117). This, of course, is regulated by the high ranking of H-COMPL (as well as *STRAY FEATURE) in VO-languages.

The main topic of chapter 3, that is, regular OS, builds upon the discussion put forward in the previous chapter, namely, that there are two types of OS: short OS triggered by the φ-features on the verbal root V that may cross the VP-adverbials, as shown in the derivation in 4 below, and regular OS triggered by the case features on the light verb v that may cross sentential adverbs, as shown in example 5 (= Broekhuis' 1 and 2, respectively, p. 119).

1. (4)

   

2. (5)

   

The requirement that a shifted object be preceded by the main verb in the Scandinavian languages is generally referred to as Holmberg's Generalization (HG). Given the D&E approach brought forth here, any form of HG cannot logically be part of the generator (=C_HL), as thoroughly discussed in section 3.1. This outcome is somewhat predicted if, as Broekhuis assumes, earlier OS and scrambling are essentially the same movement type. Since Dutch/German scrambling apparently does not obey HG, and assuming that Scandinavian OS is essentially the same operation, HG is licensed outside of GEN. The remainder of the chapter elaborates on how HG should be split up and distributed among violable interface constraints at EVAL. As an example, consider Broekhuis' observation (from section 3.2.3) that regular OS is not allowed in complex tense constructions in Icelandic, whereas it is allowed in their Dutch/German counterparts. To capture this contrast with constraints, Broekhuis concludes that in Icelandic, the constraint H-COMPL outranks both EPP(φ) and EPP(case), whereas in Dutch/German, the ranking is the opposite (= Broekhuis' 140, p. 202).

1. (6)

   

The analysis developed in chapters 2 and 3 thoroughly examines the validity of the long-held standard minimalist claim that certain universal principles are truly inviolable to determine what core components of the theory should be associated with GEN in the D&E framework. Cross-linguistic variation is modeled then by the ranking of a finite number of violable interface constraints.

Chapter 4 serves a two-fold function: First, it addresses new questions that arise as a result of the D&E approach to OS in Germanic languages. For example, in Danish and Swedish there are now reasons to assume that short OS “pushes up” the verbal root V into v/Asp. In contrast (as seen in section 3.2.1), in instances of regular OS this movement is blocked in these languages when V-to-I movement does not apply. This, of course, raises a question concerning the ranking of EPP(φ) and EPP(case), especially since the φ- and case features are also involved in triggering the movement of the subject. Second, Broekhuis takes up the task of discussing OS/scrambling's relationship with the three other types of movement in generative literature: head-movement, A-movement, and A'-movement. Concerning the first issue, Broekhuis demonstrates in section 4.1 that while Danish pushes the verbal root up into v/Asp, Swedish does not allow the verb to raise to I. To account for this dichotomy, Broekhuis suggests that the constraint NOLEXM, which disfavors movement of the main verb (due to its ability to assign θ-roles), is ranked between EPP(φ) and D-PRONOUN.

This analysis has far-reaching implications for the constraints and their subsequent rankings in other Germanic languages, such as Dutch, which Broekhuis addresses in section 4.2. Here there are actually separate EPP constraints for the [gender], [number], and [person] features. Finally, in section 4.3 Broekhuis dispels any direct connection between OS/scrambling and A'-movement. He asserts that these associations are

The topic of predicate movement serves as the main topic for chapter 5. In this chapter, Broekhuis explores the following two axioms (his example 4, p. 265).

1. (7)

   1. a. Axiom I: A predicate agrees in φ-features with the DP it is predicated of.

   2. b. Axiom II: V agrees in φ-features with its internal argument or the subject of its internal argument (when the internal argument is a small clause or an infinitival complement of the Subject Raising of the ECM-type).

Here the aim is to test the validity of this axiom under the assumption that the constraint EPP(φ) triggers short OS, as well as predicate movement. In this chapter, several different constructions from various languages are considered, such as locative inversion in English, predicate movement in Dutch, movement of the so-called verbal modifiers in Hungarian, and l-particle fronting in Bulgarian. In conformity with the D&E framework discussed and developed here, Broekhuis suggests the hypothesis in 8 below (5, p. 265) can also be upheld.

1. (8) Null hypothesis: If A and B agree in φ-features, both A and B can be the goal of unvalued φ-features on some higher head H, and, consequently, be a candidate for internal merge with H.

Although analyses proposed in this chapter are somewhat speculative, Broekhuis provides a solid starting point for future discussions and analyses of these data within the D&E framework. This book concludes with chapter 6, which offers a summary and conclusion of the work as a whole.

3. Critical Evaluation

This monograph represents almost a decade's worth of careful thought and research on the part of Hans Broekhuis, which, in my opinion, is evident in the depth of the data and argumentation presented throughout this book. What impresses me most with the D&E approach is its parsimony in restricting the candidate set generated by GEN and the finite number of constraints necessary to arrive at a descriptively adequate account of OS/scrambling in Germanic languages. There are, of course, decided advantages that the D&E framework has over main-stream minimalism. In my opinion, two such advantages that come to the forefront are the following. First, as Broekhuis himself notes at the beginning of chapter 5, the constraints in the D&E framework rival interface filters and conditions of the P&P framework. The filters that do exist in the P&P framework are quite often too rigid to account for the full gantlet of parametric variation evidenced in languages, whereas the constraints of the D&E approach are more efficient at this task. The D&E constraints allow for a more thorough understanding of what it means to be an “interface condition” with the MP. The D&E framework gives LF and PF an identifiable face and mission. Second, from an OT-theoretical perspective, the D&E framework significantly improves upon the issue of ineffability through the (significant) restriction of the GEN candidate set.

There are, of course, a few areas where the D&E framework can be strengthened and improved upon. The first of these is clearly delineating the advantages and overall theoretical design specifics that this framework has over the hybrid LFG-OT approach developed by Bresnan and others almost a decade ago. In some areas, the difference between the programs is fairly obvious (for example, one (LFG) is purely representational, while the other (MP) consists of both derivational and representational components). In other domains the contrast is not so obvious. For example, at what point in the derivation, that is, when GEN is creating (or generating) the candidate set, does EVAL get to see the candidate set? Does this happen at the strong phase level or can this wait until the entire representation has been constructed? Addressing issues and concerns such as these in the future will only improve this framework.

Second, it may indeed be possible to reduce the D&E machinery attributed to GEN even more. Under such a view it would only consist of one operation: (RE)MERGE rather than the MOVE and SELECT C_HL constraints that currently exist. Stroik (2009 and earlier work) has argued that the MERGE versus MOVE (INTERNAL MERGE) distinction is perhaps superfluous and that all that is required is iterative applications of (EXTERNAL) MERGE. Such a model of movement invokes the notion of a push-chain (rather than an attraction-based pull chain as is commonly assumed in most versions of the MP) and could easily be applied to the D&E framework, especially with its elimination of any sort of Numeration. Sappir (2011) has recently proposed a model that includes this repel-based approach to syntactic displacement, which, as far as I can tell, could be easily adopted by the D&E framework.

Third, although Broekhuis makes remarks about AGREE, it is somewhat unclear exactly how long-distance (that is, nonsisterhood feature checking) works in this framework. As suggested by Hans Broekhuis (p.c.), perhaps all of this is made implicit in the following manner: The Probe-Goal relation is absolutely necessary to make EPP(x) relevant for the evaluation to start with, given that the Goal is (x). Therefore, valuing is independent of this constraint. The ranking *MOVE >> EPP(x) indicates that Agree has taken place and nothing extra needs to take place. Restricting ourselves for the moment to A-chains and simplifying things enormously, the ranking *MOVE >> EPP(x) may be interpreted as “making the θ-relation visible in the output,” and the strong ranking as “making the Probe-Goal relation visible in the output” (Hans Broekhuis, p.c.). Of course, this is a cursory overview of Agree in this system, but in the simplest case this would be more or less the best way to implement this relation in the D&E framework.

There are some other issues with the data (such as the very complex relationship of Quantifier Raising in Icelandic and its relation with A- and A'-movement phenomena; see Svenonius 2000) that are more intricate and difficult than presented in this text. Such criticisms could, however, be leveled against any work on OS/scrambling in West Germanic due to the controversial and robust literature existing on the subject. Overall, this work and the D&E framework presented and expounded upon throughout show a great deal of promise for reasons I have outlined above. In the end, a hardliner approach as being exclusively either a lumper or a splitter is simply too strong of a stance to maintain. Given the complex nature of language and the level of its often detailed variation, it appears that certain aspects of language can be approached from a lumper's perspective, while others are better analyzed from a splitter's point of view. The D&E framework at this stage of its development has, in my opinion, been successful in incorporating aspects of both the MP and OT into its machinery and by doing so has forged a promising future for itself that will benefit from the research endeavors of other scholars.