There is a vast literature on all of the topics discussed in the past seven chapters, much of it demanding a deeper technical background than I'm assuming for people using this textbook. The following sources should, however, be largely accessible to readers who've worked their way through all of the preceding material. The bibliographies of these sources will then offer further routes to still more advanced material. Full bibliographic details on all of these suggested readings are included in the list of references at the end of this volume.

Chapter 1

The material covered in the first chapter, which is intended to assist the reader in developing a sense of the critical patterns in data from the point of view of syntacticians, is more or less standard in syntax textbooks; a particularly nice example is Robert Borsley's Syntactic Theory (1993). But probably the most comprehensive explanation of what syntacticians are really after, and how they structure their inquiries to lead to the kind of elegant, far-reaching generalizations that constitute the Holy Grail of linguistics, is to be found in Andrew Radford's Transformational Grammar: A First Course (1988).

Chapter 2

The primary technical tools in this chapter are the decomposition of category labels into highly structured complex symbols utilizing feature names and values supplied for those names, and statements of constraints, using the format of context-free phrase structure rules, on the form of possible two-generation trees in terms of feature identities which must be satisfied in order for a given tree to be legal.

The treatment of categories as complex symbols was first introduced into theoretical discourse very early in an important but generally neglected paper by Gilbert Harman (1963), which provoked a rather hostile response by Chomsky (1964), who subsequently reintroduced the use of features without reference to Harman's prior work, first at the level of lexical categories (1965) and subsequently at the phrasal level (Chomsky 1970: 207–208), where the resistance to such treatment is attributed to the American structuralists of the 1940s and 1950s – despite the fact that the decomposition of phrasal categories into a category type and level of phrasal complexity, which anticipated Chomsky's own use of complex symbols in his 1970 paper, was first proposed by just such an American structuralist, Zelig Harris, in work published in the 1940s and early 1950s.

There is a long tradition in syntax textbooks whereby the early sections (or chapters) present a particular ‘take’ on the key properties of human language (e.g., its unboundedness, the fact that speakers can pass intuitive and more or less definitive judgments on the well-formedness as stand-alone sentences of arbitrary strings of words, and so on), followed by an overview of fundamental data structure that are said to best capture these properties, followed by a series of chapters applying those data structures to a range of natural language data. Along the way, technical refinements are introduced, attempts are made to formalize the often informal statements of descriptive machinery given at the outset to jump-start the discussion, and the discussion increasingly becomes focused on the content of the theoretical framework, with natural language data used as points of departure for exploring that content. Every such framework seems to have its own ‘set piece’ examples which demonstrate its explanatory reach, its ability to capture apparently profound generalizations with a minimal number of additional assumptions. The virtues of this narrative organization are obvious: the point of any science is to capture the behavior of its objects of interest as parsimoniously as possible, which in turn requires an analytic toolkit to capture the generalizations that represent that science's discoveries about its domain of inquiry. In order to say anything useful about that domain, students must first acquire a basic working familiarity with those conceptual tools by applying them to simpler phenomena (typically using optimally simple or idealized data sets) and progressively refine and expand their mastery of the framework by tackling increasingly challenging or even open-ended problems. So far, so good.

But this kind of storyline faces a certain kind of risk: the text becomes in effect a kind of recipe book of stock analyses, with large chunks of the thinking that went into these analyses presented as faits accomplis, which students are expected to internalize and extend to new data. The result is a heavily ‘top-down’ presentation of syntax, making it a matter of mastering a typically complex set of technical tools, specialized notations, and axioms.

The topicalization construction discussed in some detail in the previous chapter is only one of a wide variety of unbounded dependency phenomena in which a filler must be linked to a gap site. One can cluster these phenomena into two major groups, on grounds we discuss below, with a third group falling somewhere between the two, showing certain properties of both. For several decades all of these constructions linking fillers to gap site have been regarded as embodiments of a single connectivity mechanism – a view allowing a considerable simplification in syntacticians’ picture of the licensing conditions on natural language grammars. But what is the basis for this assumption?

The very earliest form taken by the argument for a unitary filler/gap mechanism shared by all unbounded dependencies was based on a particular set of constraints, or restrictions, to which all constructions embodying that mechanism were sensitive. In a very influential paper, Chomsky (1977) argued along these lines for an approach in which a single movement operation underlay all types of extraction phenomena. In the previous chapter, we've seen some instances of these constructions and noted that in certain languages they all display the same grammatical behavior in marking the intermediate substructures between fillers and gap sites in some special manner. We now look more systematically at the form of these various constructions, prior to a survey of the supposed syntactic constraints that they were thought to obey. The theoretical development of this set of constraints, and the attempt to derive them as effects from ever more abstract and general principles, is arguably the driving force behind most of the developments in syntactic theory from the mid-1970s on. We conclude with a review of the empirical status of the constraints themselves, and consequences of more recent research in this domain for syntax as a field of inquiry and as a component of the grammars of human languages.

Extraction Constructions in English

Fillers in Topic Position

The constructions we consider in this section are similar to topicalization in a crucial respect: they display a constituent at the left edge of the clause which is linked to an arbitrarily distant gap site.

The auxiliary dependency analyzed in the preceding chapter takes the form of a complex network of distributional facts emerging as the consequence of a simple set of purely lexical properties. Both (i) the relative order of auxiliaries and so-called ‘main verbs’ (a notable misnomer, in our terms, since ‘main verbs’ in sentences with auxiliaries are actually deeply buried), and (ii) the morphology of clauses containing auxiliaries are due to the selectional properties of individual words, and the characteristic existence of gaps in the lexicon, along with a requirement that the top hierarchical level of English sentences be finite. Given these properties of individual words, the independently motivated featurematching requirements of the grammar, as laid out in the preceding chapters, automatically yield the correct morphosyntactic dependencies with no more effort than is required to specify that talk requires a PP complement and discuss an NP. This demonstration offers strong support for the general approach taken so far, but it would be reassuring to find that there are other, unrelated phenomena which can be satisfactorily described using the same machinery. In this chapter, we briefly survey several grammatical patterns in English and a range of other languages which can be captured rather simply using the same technology already implemented in the auxiliary analysis presented in the preceding chapter.

Case and Agreement

Case

The first of these patterns, referred to as case, identifies a generally systematic marking of the dependents of a head which often correlates with what would be called grammatical relations (subject/object/indirect object/adjunct), but which typically have additional semantic content; thus we talk about ‘dative’ case, associated with a beneficiary role, corresponding to Robin in I gave Robin a book (although this case is not realized in the form of English words) in which the transfer of some benefit to the individual marked with dative morphology is conveyed (though dative case may show up in nonbenefactive contexts as well, as we'll see below), or the possessive, as in Where is my book?, where possession is conveyed by my. In some languages, such as English, case is vestigial; although its distant ancestor Anglo-Saxon had a rich case system, reflecting the grammar of its own pan-Germanic ancestor and, ultimately, the ancestral Indo-European proto-language, English has lost almost all vestiges of this system.

A little history, to begin with: the phenomena that come under the heading ‘English auxiliaries’ were instrumental in bringing about the contemporary era in syntax, and, beyond syntax, in linguistics as a field. In Chomsky's watershed 1957 monograph Syntactic Structures, these phenomena were perhaps the key setpiece for the pivotal argument in the book that context-free phrase structure rules, of the sort that we first formulated to capture basic phrasal constituency in English, get crucial facts about the language wrong. Chomsky's point was not that such rules cannot summarize our observations about what we find in English, but rather that, implicitly, they deny the existence of a certain kind of pattern that can in fact be shown to exist, and therefore are wrong, not about the data, but about recurrent patterns in the data. The new kind of rule that Chomsky proposed to deal with this difficulty (a more user-friendly version of a kind of mechanism that certain logicians and linguists of the preceding generation (in particular Emil Post and Zelig Harris) had devised and experimented with) had the apparent virtue of being able to express these patterns easily. A good deal of the development syntax has experienced in the last quarter of a century reflects a somewhat belated counting of the costs incurred by Chomsky's innovation, and a gradual realization that there were far simpler, more economical solutions to the problem he pointed out with empirical coverage considerably better than what he had proposed in 1957.

The particular virtue of Chomsky's approach, then, is not the particular solution he advanced, which has become progressively less attractive as time has passed (even independently of framework-internal changes in the specific theory he inaugurated), but rather the enhanced scope of the linguist's intellectual responsibilities that it assumed. Recording the facts correctly no longer had the status of a sufficient activity, even with a convenient classification and organization of those facts; what was now essential was motivating those facts, by showing how they were manifestations of underlying patterns and their interaction. One could characterize most previous work in linguistics as at best proto-scientific, in the sense that recording and classification are crucial to the actual business of science, but are not themselves that business, which is to identify a domain of phenomena and discover the general principles governing the behavior of the phenomena in that domain.