Cedric Boeckx's Bare syntax is a very welcome new addition to Minimalist theory. It combines an accessible introduction to core concepts in current Minimalism with a substantive new proposal on the status of projections and locality. This combination of an introduction with a new proposal is, I believe, very characteristic of Boeckx's writing and is one of its biggest strengths. Highly programmatic in character, the book combines many different ideas and puts them together in an intriguing new way. More specifically, it discusses three domains typically thought to be unrelated: (i) locality of movement, (ii) labels, and (iii) extended projections. The central claim is that restrictions on locality, the need for unique labels, and the general template for extended projections all follow from the need to provide unambiguous input to the interface systems SEM and PHON.
Chapter 1, ‘Preliminary considerations’, sets the scene for what is to come. It provides an insightful introduction to the essence of Minimalism, without getting into too many technical details. Boeckx emphasizes the perfection and optimality of the language faculty with respect to the interfaces and offers some initial thoughts on locality, previewing the need for a unified approach to locality of selection and locality of movement. With respect to locality of selection, he focuses on how the expansion of functional projections complicates a simple sisterhood-based approach to selection. For example, a verb cannot select an object directly as its sister because the noun phrase (NP) is ‘buried’ within a number of intervening functional projections. This chapter also addresses locality of movement, summarizing three types of existing answers to the fundamental question of why islands exist. These are: (i) islands are a matter of computational efficiency (thus they exist because of narrow syntax considerations), (ii) islands exist because of the restrictions imposed on syntax by the external systems (conceptual-intentional and sensory-motor systems), and (iii) islands exist because of reasons outside of the faculty of language in the narrow sense (such as processing or memory limitations). The view of islands that Boeckx develops later in the book links islandhood to the conditions imposed by the interfaces.
Chapter 2, ‘Outline of a general theory of locality’, establishes a number of equivalences and parallelisms in syntax. The first one concerns Merge and Move (or, in more current terms, External Merge and Internal Merge). Boeckx follows Chomsky (Reference Chomsky and Belletti2004) in assuming that there is only a single operation Merge, whose input could be either two disjoint objects or two objects one of which is a subpart of the other. He then takes this External/Internal Merge equivalence even further in arguing that the product of Merge is equivalent to the product of Move. This leads to the equivalence (33) given in (1).
(1) The Product of External/Internal-Merge Equivalence
Projection and Chain are symmetric objects.
To establish this equivalence, Boeckx borrows from ‘more advanced sciences’ (34), such as physics, and shows that chains and projections can be unified since they are (i) commensurable, (ii) permutable, and (iii) relatable by a symmetry transformation. Commensurability of projections and chains becomes apparent, he argues, if we think of the distinction between minimal, intermediate and maximal projections as corresponding to the distinction between thematic, intermediate and final positions in a chain created by movement. The evidence Boeckx gives in favor of chains and projections being permutable is twofold: a chain can become a projection, and a projection can become a chain. The former is the case if a moved element projects. According to Donati (Reference Donati, Cheng and Corver2006), this is what happens in free relatives, in which the moved wh-element (e.g. what in I eat what John eats) moves and projects. Boeckx's evidence for chains becoming projections is a little less straightforward. It involves ‘reprojection’, i.e. cases in which a given element, instead of undergoing movement, can simply project a new label in situ. He examines two such instances: Hornstein & Uriagereka's (Reference Hornstein, Uriagereka, Epstein and Seely2002) analysis of strong quantifiers, which, they argue, reproject as transitive predicates of sorts in covert syntax, and English prepositional experiencers, which, according to Kitahara (Reference Kitahara1997), reproject as Determiner Phrases in covert syntax. Interestingly, both of these cases of reprojection are covert, which raises the question not only whether reprojection is limited to the covert component but also how it can be restated in the current Minimalist model where the overt/covert distinction is dispensed with and covert movement operations are captured via Agree.
Chapter 3, ‘Unambiguous Merge’, starts out by examining the basic properties of the two interface systems: SEM (the interface with the conceptual-intentional system) and PHON (the interface with the articulatory-perceptual system). It lists the minimal interface requirements imposed by PHON and SEM on narrow syntax. To meet the needs of PHON, hierarchical syntactic representations must be at the very least converted to linear strings and assigned a prosodic structure (which Boeckx assumes is done in a way that reflects syntactic hierarchy). With respect to the requirements of SEM, Boeckx departs from the views of semantics that rely on different modes of composition (such as Functional Application and Predicate Modification in Heim & Kratzer's Reference Heim and Kratzer1998 approach) and adopts Pietroski's (Reference Pietroski2005) system, which takes concatenation to be the only mechanism necessary for the interpretation of complex expressions. Such a system makes syntax and semantics isomorphic. Given that Merge is the only concatenation mechanism available in narrow syntax, it is to be expected that there will be only one corresponding semantic mechanism. Next, Boeckx turns to the discussion of what properties Merge needs to have in order to meet these interface conditions in an optimal way. He argues that even though Merge is an inherently symmetric operation, the product of Merge has to be asymmetric in order to be interpretable at the interfaces. This asymmetry comes in the form of the label. Boeckx proposes that labels render the structures created by Merge unambiguous to the interfaces, which raises the interesting question of whether both interfaces need labels. Labels are said to make the mapping to the SEM and PHON interfaces maximally efficient – an idea captured by the ‘Quick Edge Detection conjecture’ (91), given in (2).
(2) Quick Edge Detection (QED) conjecture
To ensure efficient, optimal mapping, processing decisions must be made as quickly as possible. Quick edge detection contributes significantly to mapping efficiency.
The next question concerns the identity of the label. Which of the two merged elements becomes or determines the label? To answer this question, Boeckx relies once more on the unification of Merge and Move, and proposes that the label is the element that has its uninterpretable feature (uF) valued via Merge (96).
(3) Probe-Label Correspondence Axiom (PLCA)
The label of {α, β} is whichever of α or β probes the other, where the Probe=Lexical Item whose uF gets valued.
An intriguing consequence of this proposal is that it gives rise to what might be called ‘sideward Agree’. We know that feature valuation requires Agree, so using Agree to motivate Merge or to determine the identity of labels in structures created by Merge seems to require Agree between two disjoint objects. Another interesting issue is raised by cases in which both objects undergoing Agree have some unvalued features valued. If such cases exist in Agree between disjoint objects, we should expect either object to be able to determine the label.
Boeckx also makes a substantive proposal regarding adjuncts in this chapter. He takes adjunction to be a more basic operation than complementation in that it creates a set with no label (as also proposed by Hornstein & Nunes Reference Hornstein and Nunes2008). Boeckx advocates a ‘later merge’ theory of adjunction, according to which adjuncts are attached to elements already present in the derivation. The new object has no label (or, perhaps more accurately, the label of the existing object remains unchanged); thus adjuncts are not part of the projection/chain-representation. This predicts that adjuncts should not participate in any syntactic relationships. This is certainly not true for all cases of adjunction, as Boeckx rightfully admits. While adjuncts are more restricted in their movement possibilities than complements, they are not generally barred from movement and can enter into covert dependencies or intervene to block movement of other elements. For Boeckx's proposal to be feasible, adjuncts with these properties would have to be treated as quasi-arguments of sorts.
Chapter 4, ‘Cartographies and the locality of selection’, pursues the idea that sequences of projections – which Boeckx refers to as cartographies, building on the vast literature on the topic – are organized according to the same principles as simple projections and chains. The principle involved in all three cases is what might be called a generalized X′-template, specifying three distinct levels of structure. Extended projections in Grimshaw's (Reference Grimshaw1991) sense involve X′-like structure in that they also consist of three layers: a lexical layer (corresponding to X0), a first level of functional layer (corresponding to X′), and a maximal level of functional layer. For more ‘exploded’ extended projections, this implies division into three different types of projections. For example, Boeckx follows Rizzi (Reference Rizzi and Haegeman1997) in decomposing the Complementizer Phrase into a sequence of four projections (ForceP, TopicP, FocusP, FinitenessP). The topic/focus layer is iterative and optional, and thus corresponds to the (also optional and iterative) intermediate layer in an X′-projection. Since FocP is iterative, the ban on multiple foci in languages like Italian requires an independent explanation.
Chapter 5, ‘Islands and the locality of chains’, extends the proposal to locality on chains. In most general terms, it deals with three issues: immobility of moved elements (in particular subjects), island effects (in particular Condition on Extraction Domains (CED) effects), and island repair strategies. The core constraint on chains (167) is given in (4).
(4) An element can only move to a single feature-checking site.
Only chains that meet the condition in (4) are QED-compliant in that they provide unambiguous instructions to the interfaces. The interfaces ‘stop scanning chains once the first unambiguous checking site is reached’ (174). Generally speaking, a checking domain for Boeckx consists of two heads, one containing a [−F] and the other a [+F] specification of some feature F. Construed in this way, a checking domain can contain only one movement/checking site, the location of which is indicated with an underscore in (5).
(5) [H10[−F] [ __ [H20[+F] ]]]
Boeckx argues that the principle in (4) is all we need to derive locality effects. This is a bold statement and the bulk of the discussion in the rest of the chapter goes into justifying it. First, he shows how the (relative) immobility of subjects (as opposed to objects) follows from (4). Since moved subjects have already entered a feature-checking relationship, they are frozen for further movement. Boeckx further argues that his understanding of checking site allows for two ways in which subjects can undergo further checking. He takes the checking position for subjects to be as given in (6).
(6) [Fin10[+T] [ __ [T0[−ϕ]]]]
Checking will fail if the featural content of either Finiteness (Fin) or Tense (T) is altered (either via the loss of phi-features on T or via the loss of the T-feature on Fin), thus freeing the subject for further movement. He proceeds to argue that the various strategies that languages resort to in order to move subjects (such as the use of null complementizers, resumptive or expletive pronouns, clausal pied-piping, and anti-agreement effects) correspond to one or the other of these two types of feature loss.
In the second part of chapter 5, Boeckx considers islands, focusing on CED effects. The reasons for island violations are similar to the reasons for labels and cartographies, i.e. avoidance of ambiguity at the interfaces. He thus derives the Subject Condition part of the CED from the QED-requirement: extraction from a moved element (such as a subject) is banned because the subject has already entered into one checking relationship and moving a part of it to another feature-checking site will violate (4). If a moved element has exhausted its chain potential, no part of it will be able to enter into new checking relationships. This makes the interesting prediction that extraction out of subjects should be possible if any of the strategies for moving a displaced subject has applied. Boeckx examines familiar evidence showing that movement out of non-displaced subjects is possible; however, it would also be interesting to see if there are any cases in which movement out of, for example, non-agreeing subjects is possible as well.
In the last part of this chapter, Boeckx turns to island repair strategies, providing a unified account for why resumption, sluicing, and pied-piping can remedy islands. He starts with an analysis of island obviation effects via resumption and ellipsis, building on his earlier work on resumption and Wang's (Reference Wang2007) analysis of sluicing as involving a null resumptive pronoun in the ellipsis site. The upshot of the discussion in this chapter is that islands are a result of ambiguous chains. The island repair strategies discussed in this chapter all have the effect of dividing an otherwise ambiguous chain into two non-ambiguous chains. For example, resumption rescues island violations because a resumptive pronoun introduces its own chain, and the moved element does not enter into a checking relationship till its final position. Boeckx extends this account to wh-in-situ elements, which are well known (at least in some languages) to be insensitive to islands. If we follow Watanabe (Reference Watanabe1992) and think of a wh-in-situ element as a combination of a null operator and an indefinite part (which Boeckx reanalyses as a resumptive pronoun of sorts), the parallelism becomes quite obvious. For Boeckx, pied-piping is a mirror image of resumption. The moved element, instead of being completely dissociated from the island due to the presence of the resumptive element inside the island, drags the entire island with it. Thus it does not enter into new checking relationships.
In conclusion, Boeckx's Bare syntax offers a fresh and stimulating perspective on core issues in current syntactic theory, and proposes a unified approach to a number of seemingly non-unifiable phenomena. Its empirical strength lies in breadth rather than depth. A reader looking for a detailed in-depth examination of one empirical domain (or a small set of related ones) might leave a little disappointed. A reader looking for inspiration on how to connect a range of relatively well-known phenomena in intriguing new ways will leave quite inspired.
