A Companion to Chomsky. Группа авторов
1960s, yielded a research methodology whose core features guarantee quick and fruitful syntactic research.
The founding dictum of generative linguistics is that linguists should study grammars, where a grammar of a particular language L is understood as a set of rules for generating the sentences that constitute L (Chomsky 1955 et seq.). From the late 1950s onward, Chomskyan generative grammar has also adopted a rationalist conception of language (Chomsky 1958, 1959), according to which “a grammar of a particular language must be supplemented by a universal grammar [our emphasis] that accommodates the creative aspect of language use and expresses the deep‐seated regularities which, being universal, are omitted from the grammar itself” (Chomsky 1965, 6). This engenders a linguistic research program with two main goals: I. to specify the rules that generate all and only the grammatical sentences in a given language L (we return to the concept of grammaticality momentarily), and II. to determine which of these rules belong to the universal grammar and which are specific to L.
These two goals are directly reflected in the research methodology employed by generative linguists from the 1950s until today. The first goal is reflected in the choice of data‐type used for generative linguistic research, while the second is reflected in the methodological outlook adopted by generative researchers.
To accomplish goal I., a linguist will (ideally): (i) hypothesize a grammar for L and then (ii) test to see if it correctly generates all and only the grammatical sentences in L. To successfully carry out this test, a linguist must know beforehand the grammatical status of sentences through an independent test to avoid introducing circularity into the methodology described above. This independent test comes from acceptability judgments. Acceptability judgments are reports provided by a native speaker of their spontaneous reaction concerning whether a particular string of words (with an intended interpretation) is a possible sentence of their language. Generative linguistics has used acceptability judgments as its primary data source since its inception (Chomsky 1955, 1957), treating these judgments as a valid and valuable source of information about the grammaticality of a sentence. In other words, generative linguistics has always assumed that the reaction triggered by exposure to a candidate sentence of one's native language can help determine whether the sentence can or cannot be generated by that language's grammar.2 In addition to this tacit acceptance of acceptability judgments as a valid data‐source, Aspects of the theory of syntax (1965), Chomsky's most cited book on linguistics, explicitly defended their use, arguing that no adequate alternative method of probing linguistic competence was available (at the time) or would likely be discovered, and that, on the off‐chance that an alternative method was developed, acceptability judgments will still be required as a benchmark against which to test the alternative method's efficacy (p. 18–21). Chomsky therefore highlights that acceptability judgments are the most easily obtained source of ungrammatical sentences, which are indispensable for conducting the everyday task of generative research, namely developing accurate characterizations of grammars.3
Accomplishing the second main goal of generative linguistic research has been regularly delayed by Chomsky's periodic refinement or revision of his conception of the universal grammar and its relation to language‐specific grammars (Chomsky 1970, 1977, 1981, 1986b, 1995). Despite periodic conceptual changes, the methodological outlook used for pursuing this goal has remained constant. One aspect of this outlook is the adoption of a working method in which productive patterns observed in one language are presumed to be present in all languages until empirical evidence proves otherwise. Because discovering the properties of the universal grammar is more pressing than discovering the rules of the ancillary language‐specific grammar from a rationalist perspective such as Chomsky's (as the universal grammar is innate whereas the language‐specific grammar is learned), another aspect of this outlook is its focus on linguistic regularities, or “core language” (Chomsky 1986a, 147).
In sum, the theoretical commitments emplaced by Chomsky in the late 1950s and early 1960s engendered a research methodology in which informally collected acceptability judgments are used to make generalizations over linguistic regularities. Considering that acceptability judgments are reliable, relatively unequivocal (see the references in endnote 3), and are quickly obtained (especially if a linguist is collecting judgments from herself and/or a small group of consultants), and that the rationalist principles underpinning the generative paradigm has given linguists warrant to make universal claims from studying one language that they know well (usually their native tongue), it is unsurprising that this methodology yields new data, generalizations, and analyses at an extremely rapid rate. Equally, it is unsurprising that its initial application in the early days of generativist syntax to what was then an uncharted empirical landscape of syntactic regularities resulted in most of the profound – and therefore the most enduring – discoveries of generative syntax.
4.3 Generative Syntax Through the Lens of Nonlocal Dependencies
Although generative linguistics is predominantly a syntax‐focused program (see Jackendoff 2002 for criticism of this fact), the methodology described in the previous section is intended for use in all linguistic subfields (see Chomsky and Halle 1968 for a famous application of the methodology to phonology). Its enduring association with syntactic research stems from the fact that Chomsky first employed the methodology to study the regularities of English syntax in Syntactic Structures (1957). In addition, Syntactic Structures also set the research agenda for precisely which syntactic regularities linguists would focus on in the early period (the 1960s and 1970s) of generative theorizing. From a broad, theory‐neutral perspective, the regularities in question can each be described as instantiating a nonlocal dependency. One example of a nonlocal dependency is the relationship that obtains between every Englishman and his in (1): the specification for his depends on the specification for every Englishman (i.e. Gerald takes pride in Gerald's garden, Norman takes pride in Norman's garden, etc.), even though these phrases are separated by the string of words takes pride in.
1 (1) Every Englishman takes pride in his garden.
This concrete notion of a nonlocal dependency can be extended to include a slightly more abstract case, such as the dependency into which the nominal expression the cake enters in (3). In (2), this expression occupies a position immediately following the verb. Because this position is the designated object position in English, the cake is understood as the sentence's object, i.e. the thing being eaten. In (3), however, the cake is still understood as the object of the sentence, despite being separated from the position immediately following the verb by two words (namely, was and eaten). In this configuration, one may characterize the cake as entering into a nonlocal dependency with the position immediately following the verb, which is represented by the symbol “Δ” in (3) and is known as the gap (or an empty category).
1 (2) Ashley ate the cake.
2 (3) The cake was eaten Δ by Ashley.
Influenced by ideas developed by Harris (1952, 1957), Chomsky in Syntactic Structures (1957) introduced the idea that certain sentences are derived from other sentences via rewrite rules over phrase markers (a special type of derivation tree whose termini are adorned with the atoms of syntactic analysis, namely words and morphemes), which he called transformations. One such transformation is the passive transformation, which converts the phrase marker for an active sentence such as (2) into its passive counterpart in (3). Although Chomsky did not originally view the output of a transformation as instantiating a nonlocal dependency,4 the way in which his transformational analysis focused generative linguists' attention on nonlocal dependencies from the outset of the generativist enterprise is clear, as transformations suggest that some nonlocal connection – in this case, a derivational history – obtains between the two positions of the cake in (2) and (3). This early focus on nonlocal dependencies