Phrase Structure Composition and Syntactic Dependencies (review)

The content of this book may be viewed in at least two ways. Most straightforwardly, it is an exposition of tree adjoining grammar (TAG) as the formal basis for a theory of syntax. TAG was first defined in Joshi et al. 1975 and has been developed since by Joshi and a number of his students and colleagues, the author having been one of the former. While those with an interest in formal grammars and computational linguistics know TAG from a number of papers that have been written on its formal properties, this book is aimed at syntacticians who do not necessarily have any prior interest in (or belief in the relevance of) the position on the Chomsky hierarchy of the grammar formalism they or others employ. From another angle, this book may be seen as an interesting and original exploration of the notion of locality in syntactic theory—the empirical pervasiveness of such a concept, how it can be reconciled theoretically with unbounded movement, and, most centrally, the claim that it is a fundamental property of the underlying grammatical system.

After a brief preface, the book consists of six chapters. Ch. 1 accomplishes two goals. First, it situates TAG within the history of the generative paradigm—a discussion of more than just historical interest, since both TAG and the minimalist program are illuminated by an understanding of their mutual relation via the generalized transformations of Chomsky 1955. Second, it outlines how the formalism works, in particular the two ways in which the finite tree structures called elementary trees can be combined: by substitution and adjunction. The first may be thought of as a way of inserting a subtree into an unexpanded node at the frontier of another tree; the latter allows a subtree to be inserted at any node within another tree. Both operations introduce the possibility of unbounded recursion into the grammar, but in importantly different ways that are explored in the remainder of the book. In this chapter Frank discusses what he calls the fundamental tag hypothesis: that every syntactic dependency is expressed locally within a single elementary tree. For example, dependencies such as the one between a ‘raised subject’ like John and its trace in the lowest object position (or, more theory-neutrally, between the NP and the verb of which it is an argument) in John seems to have been expected to be elected t, or the one between a wh-phrase and its trace in Who do you think that they will say will be allowed t to leave?, are restricted by the nature of the formalism itself to a particular domain (the elementary tree); the apparent unbounded nature of these dependencies can only arise through subsequent adjunction operations introducing one or more subtrees between the head and the tail of the dependency.

Having established that elementary trees constitute the domain of locality within the formalism, F argues in Ch. 2 for a particular definition of this domain as the extended projection of a lexical head (Grimshaw 1991) and discusses the formal details of how this can be implemented. Ch. 3 applies the theory as developed up to this point to one well-known and long-discussed syntactic phenomenon, NP-movement or ‘raising’. His goal is to show how a number of limitations on this construction (including the ungrammaticality of ‘superraising’, the position of the associate of expletive there, the absence of raising in nominals and (some) gerunds, and the behavior of predicate nominals) follow from the nature of the formalism itself without recourse to additional principles such as minimalism’s ‘shortest move’, ‘extension condition’, the preference for Merge over Move, and so on.

Ch. 4 explores how to characterize crucial properties of elementary trees in the absence of these principles and includes an attempt to adapt Chomsky’s system of feature-checking in a way that is consistent with the TAG framework. It also includes an extensive discussion of how to handle apparent violations of locality...