In the early days of generative grammar (the late 1950s and early 1960s), the syntactic practice of American Structuralist linguistics was equated with the newly formalized theory of Context-Free Phrase Structure Grammar (or Context-Free Grammar (CFG)). `Proofs' were then offered that CFG was inadequate for the description of natural languages, a claim that played an important role in the transformational revolution, leading to the widepsread acceptance of transformational rules and the general framework of Transformational Grammar (TG). These arguments for the inadequacy of CFGs, repeated in dozens of textbooks, gave TG the air of a rigorous science, one that people would try to apply (not very successfully, as it turns out) to everything from language teaching and computer language processing.
In one of the most important papers written in 20th century linguistics, Geoff Pullum and Gerald Gazdar demonstrated that all arguments against the context-freeness of natural language published prior to 1982 are either mathematically flawed or based on incorrect assessments of the empirical data. The reference is:
There is still some debate about who first actually did prove that CFGs are inadequate for analyzing natural languages (see Pullum 1986, 1987). Sound arguments that some natural language is not context-free are presented by Culy (1985) and Shieber (1985) (both Stanford students at the time):
The majority of linguists, however, didn't read any of the technical papers discussed by Pullum and Gazdar. Rather, they were inluenced by the informal arguments made against context-free grammars that they read in Chomsky's Syntactic Structures (a textbook for the introductory course in transformational grammar at MIT) and the highly influential review of that book published in Language (1957) by Robert B. Lees:
Both Syntactic Structures and Lees' adulatory review of it are an interesting read from the perspective of 21st century linguistics. The arguments presented by Chomsky (and strengthened rhetorically by Lees) have little force against the specific alternatives to transformational analyses of coordination, passivization, the English auxiliary system, etc. that have been developed by various scholars in the last quarter of the 20th century. In any case, the arguments one finds in these influential works all appear to be flawed in that: (1) they are more innuendo than proof (they include none of the `proofs' discussed and rebutted in Pullum and Gazdar's paper) and (2) they fail to consider alternative analyses that express the observed generalizations without transformations, i.e. in terms that would be compatible in principle with phrase structure grammar. In short, Chomsky and Lees, rhetoric aside, present no sound arguments against CFG.
The abandomnent of CFG was thus a social fact -- one not based on any scientific results demonstrating in any sense that CFGs were empirically inadequate. The same is true of the worldwide acceptance of transformational grammar. The field appears to have been ready and willing to believe that the practice of American structural linguistics could be formalized as CFG (which I do not believe was ever shown), that that practice was wrong (which was also never shown, at least not in the 50s, 60s, or 70s), and that transformational grammar was the only promising alternative to pursue (which was never, to my knowledge, even argued). This situation is reminiscent of Arthur Koestler's account of the shift to heliocentrism in 16th century cosmology:
The Pullum-Gazdar refutation of earlier arguments against the context-freeness of natural language provided part of the initial motivation for reconsidering CFGs (augmented in familiar ways, e.g. by X-Bar Theory). However, what attracted considerable interest in GPSG on the part of many linguists was the fact that it provided a simple, elegant account of (one part of) Ross's (1967) Coordinate Structure Constraint and its `across-the-board' exceptions, which had long vexed transformational grammars. This interest began with Gazdar's pioneering 1981 paper:
The complex properties of the English auxiliary system were systematized by Chomsky in the late 1950s in a way that attracted many people to TG. However, the GPSG framework, which used metarules to express generalizations thought to require transformations, provided the basis for a comparable (if not superior) CFG treatment:
As the theory of GPSG evolved, it became more modular and constraint-based (but dependent upon defeasible constraints and complex interactions thereof). The theory was codified in the 1985 book by Gazdar et al. and the treatment of coordination presented by Sag et al. (1985):
The issues surrounding the coordination of unlike categories and syncretism (e.g. of case) in coordinate structure were hotly debated in a number of papers by Stefan Dyla, Annie Zaenen, Lauri Karttunen, Arnold Zwicky, Geoffrey K. Pullum, Pauline Jacobson, Sam Bayer, Mark Johnson, Roger Levy, Carl Pollard, Mike Daniels, and others. For an overview and a recent proposal, see:
Finally, for a demonstration that GPSG, as developed in the previous works, can be compiled into a purely monotonic constraint-satisfaction system, see Shieber's (1986) paper:
In class we discussed various classes of formal grammars and the current state of the art, including Aravind Joshi's suggestion that all natural languages fall within the class of `mildly' context-sensitive languages. In this connection, we mentioned both Tree-Adjoining Grammars (TAGs) and Pollard's Head Grammars. For more information about these, please consult:
Last Updated by Ivan Sag: Mar 31 2009