A Companion to Chomsky. Группа авторов
of the early enduring discoveries of generative syntax.
In the remainder of this section, which constitutes the main body of our overview, we first outline some of the enduring discoveries made about the syntactic structures that serve as the scaffolding for nonlocal dependencies (Section 4.3.1), before turning our attention to enduring discoveries made about nonlocal dependencies themselves (Section 4.3.2). It will become clear from the discussions in these sections that natural language contains many gaps (also known as empty categories): Section 4.3.3 discusses the variety of gaps that generative linguists have uncovered.
4.3.1 Nonlocal Dependencies and Hierarchical Structure
The discovery that the establishment of a nonlocal dependency rests on hierarchical relations between words and phrases rather than on linear relations represents a watershed moment for generative syntax. This discovery was first documented by Klima (1964), who introduced the notion of in‐construction‐with (p. 297), whose converse was rebranded as c‐command by Reinhart (1976). C‐command is defined over nodes in a phrase marker. According to Reinhart's definition (p. 32), a node A c‐commands a node B if and only if the first non‐unary node that dominates A (call it C) also dominates B (see (4)). (Aside from A, B, and C, the labels decorating the nodes in (4) and (6) are arbitrarily assigned.)
1 (4)
The fact that nonlocal dependencies rest on c‐command is illustrated in the following sentences. Any in (5a) is a negative polarity item (NPI), whose use must be “licensed” by it entering into a dependency relation with a negative item such as nobody (in (5b), any is not permitted because nobody is absent; Klima 1964). Although it might seem from comparing (5a) and (5c) that nobody must merely linearly precede any for a dependency to be established between them, (5d) shows that an appeal to linearity alone is insufficient, as nobody linearly precedes any yet the sentence is unacceptable. The crucial insight here is that nobody also c‐commands any in (5a) (transposed onto (4), nobody occupies A's position and any occupies B's position), whereas nobody does not c‐command any in (5d), as (6) shows. This insight not only refined ideas about syntactic dependencies, but also conclusively demonstrates that natural language syntax is sensitive to hierarchical structure: it is a two‐dimensional phenomenon. Thus, the discovery of c‐command underscores the notion, which is implied in Chomsky's (1957) pioneering use of phrase markers, that syntactic relations between words/morphemes are predominantly (and perhaps exclusively; see Kayne 1994) hierarchical in nature, and that linear order is, perhaps counterintuitively, largely irrelevant to natural language syntax.
1 (5)a.Nobody thinks that Harriet wants any wine.b.*Somebody thinks that Harriet wants any wine.5c.*Any wine is wanted by nobody.d.*That nobody bought beer means we must drink any wine.
1 (6)
A c‐commands each boxed node, which does not include B. Therefore, a nonlocal dependency between nobody and any cannot be established.
Our understanding of the hierarchical structure on which nonlocal dependencies are instantiated has undergone substantial refinement since Chomsky's (1957) pioneering use of phrase marker diagrams. From Syntactic Structures until the mid‐1970s, words from lexical categories (e.g. nouns and verbs) and functional categories (e.g. articles such as the and auxiliary verbs) were treated dissimilarly with respect to syntactic structure. Lexical words were usually analyzed as heads of syntactic phrases (where a head of a phrase determines the phrase's category), whereas functional words were not. To provide an example, the noun (N) dog in (7) is the head of the noun phrase (NP), but the determiner (D) the is not the head of a determiner phrase. Instead, it is contained in the noun phrase.
1 (7)
Building on results obtained from the mid‐1970s through to the mid‐1980s (in particular, Emonds 1976, Jackendoff 1977, Fukui 1986, Speas 1986, and Abney 1987), Chomsky's (1986b) Generalized X′‐theory abandoned the notion that lexical and functional words are structurally distinct. Instead, functional words behave just like lexical ones: they are heads of syntactic phrases (a determiner will ‘project’ a determiner phrase (DP), for instance (8)).6
1 (8)
The distinction between lexical and functional items has a long tradition in linguistics. Consequently, the discovery that lexical and functional words are structurally equivalent (insofar as both project phrases) has endured because it eliminated a potential source for this distinction and therefore stimulated much research into precisely where, if not in terms of projecting phrases, the difference between lexical and functional items lies. Grimshaw (2000 [1991]) discovered that the crucial difference can be found in the respective ordering of lexical and functional phrases: lexical phrases are always dominated by functional ones. Moreover, there is a constrained pairing between lexical and functional phrases, with functional phrases acting as the extended projections of the lexical phrases they dominate. For instance, the extended projections of lexical N are the projections of functional D, and these share the same categorial features with N (e.g. nominal).
Grimshaw's work in turn stimulated much research into the precise make‐up of various extended projections. Certain conclusions reached in this subfield can already be viewed as “enduring discoveries,” based on the weight of evidence marshalled to support them and for their utility as diagnostics of other syntactic properties. An instructive example comes from how verbs project syntactic structure. Researchers recognized that, instead of having functional projections related to tense and aspect as the immediate extended projections of V (as in Grimshaw's original system), V is immediately dominated by (at least) one functional projection that relates to the internal thematic/eventive meaning of V. The head of this projection, which is referred to as “v” by Chomsky (1995), is broadly identified as the syntax reflex of agentivity or causation, as it selects the external argument (i.e. the agent) of an event (Larson 1988; Hale and Keyser 1991, 1993; Krazter 1994; Chomsky 1995, 2000) (9).
1 (9)
Support for the existence of v comes from the fact that its presence explains two independent enduring discoveries. The first concerns the discovery from Perlmutter (1978) that two classes of intransitive verbs – namely, unergative and unaccusative verbs – can each be associated with different syntactic phrase markers. Despite their surface similarities, unergatives have an underlying subject while unaccusatives have an underlying object. A natural structural explanation for this distinction is that unaccusative verbs have no vP projection (though see Alexiadou, Anagnostopoulou, and Everaert 2004). The second enduring discovery comes from Burzio (1981), who observed that a verb can assign a thematic role (e.g. the agent, experiencer, source, etc., of an event) to its subject if and only if its object can receive accusative case. This is Burzio's generalization. If the functional head v is responsible for introducing the subject and assigning accusative Case to object DPs, then if v is missing, then Burzio's generalization is explained.
In summary, generative research has underscored the importance of syntactic hierarchy in encoding traditional linguistic distinctions (e.g. lexical versus functional), encoding compositional meaning (e.g. agentivity), and establishing dependencies between words and/or phrases.
4.3.2 A Typology of Nonlocal Dependencies