Review of Ideology and Linguistic Theory: Noam Chomsky and the deep structure debates by Geoffrey J. Huck and John A.Goldsmith
Huck and Goldsmith’s book, Ideology and Linguistic Theory (henceforth ILT), presents a new perspective on the generative semantics distraction in the history of modern generative grammar (roughly 1967–75). During this period, a group of transformational grammarians (principally George Lakoff, James McCawley John Robert Ross, and Paul Postal) attempted to assume leadership of the field by developing a program for semantic research (called “generative semantics,” henceforth GS) based on a critique and revision of the theory of syntax in Chomsky (1965, henceforth Aspects), thus presumably in opposition to Chomsky’s own research program. At the outset, GS generated a great deal of interest and excitement, but by the mid-1970s it had more or less disappeared from sight. ILT, like other historical accounts of the period (Newmeyer 1980, 1986; Harris 1993), repeats the familiar tale of an intense and acrimonious debate between proponents of GS and Chomsky over the existence of a linguistic level of deep structure. However, unlike Newmeyer and Harris, Huck and Goldsmith argue that the debate was inconclusive and therefore cannot account for the demise of GS. Instead, they argue that external factors, and not the disconfirmation of its central claims, were responsible for the failure of the GS research program. Nonetheless, the current consensus among theoretical linguists, they claim, is that GS “was falsified,” in spite of the fact that its central claims are “by and large still accepted” (p. 92).
…significant chunks of what were evidently standard Generative Semantics analyses began to reappear in the syntax literature shortly after the movement’s demise and are now often regarded as constituting preferred solutions to contemporary problems. Indeed, the picture of grammar presented in much contemporary work, including, for example, Chomsky’s Lectures on Government and Binding (1981) and Knowledge of Language: Its Nature, Origin, and Use (1986), is similar enough in certain crucial respects to that painted by Generative Semantics in the late 1960s that one who accepts the standard story must be prepared to explain why criticisms of the latter do not also apply to the former. [p. 2]
This apparent paradox provides the basis for their conclusion that contemporary linguistic theory is governed by ideology If their view is accurate, then ILT provides a case study of irrationality in what is supposed to be a field of rational inquiry (cf. Kuhn 1962; Feyerabend 1975) and therefore is of interest beyond the history of generative grammar.
While some of the material in ILT does recreate the flavor of the period—in particular, the bits of correspondence between Chomsky and others, and the interviews with some of the most prominent researchers of the period (Jackendoff, Lakoff, Ross, and Postal) included in an appendix that makes up half of the book—the major claim of this study seems easily refutable, even if we grant the assumption that the central claims of GS were not disconfirmed (but see Newmeyer [1996] and the references cited by him for the contrary view). The basis for this claim is weak at best, including unmotivated interpretations and a general lack of solid evidence.
GS arose as a particular articulation of the Aspects model, which was the first of Chomsky’s works to discuss the interface between syntax and semantics in any detail. “The earliest recognizably GS works were attempts to fill gaps and rectify defects in the Aspects conception of syntax while preserving most of its framework” (McCawley 1995:311). Yet ILT discusses Huck and Goldsmith’s view of the 1967 GS program five pages before the Aspects program, giving a rather misleading impression of how the two research programs were related.
In chapter 2, the discussion of competing research programs is organized around sketches that are intended to identify the core propositions of the program and its basic orientation. For example, Chomsky’s initial research program, presented in The Logical Structure of Linguistic Theory (1955a, henceforth LSLT), is characterized as being “distributional” in orientation with the following core propositions:
1. The distribution of linguistic elements is determined by structural properties of the grammatical strings made up of those elements.
2. The formal (phonological, syntactic) part of the grammar is autonomous from the semantic part.
3. Structural differences among the grammatical sentences of a language can be accounted for by a set of phrase structure rules and the transformations that apply to them. [p. 14]
Huck and Goldsmith imply that LSLT is primarily concerned with the distribution of linguistic elements to the exclusion of other concerns (e.g., sound-meaning correspondences, which they designate as a “mediational” orientation). However, as Chomsky notes in his introduction to the 1975 edition of LSLT, the point of view assumed throughout was that “the grammar is to provide a means for semantic description, and should fall in place in a broader semiotic theory in which this promissory note is made good” (Chomsky 1975a:19). Moreover, “semantic criteria are essential for determining the soundness of the general approach, since the grammar is to provide the means to account for the use and understanding of language (Chomsky 1975a:19). In LSLT, Chomsky distinguishes between the appeal to meaning and the study of meaning, the latter being “an essential task for linguistics” (p. 97). Thus it is rather misleading to characterize the LSLT program as completely unconcerned with the semantics of natural language. (See also Chomsky [1955b] for additional discussion of semantics.)
The claim that the LSLT program was unconcerned with semantics appears to be reinforced by what Huck and Goldsmith list as its second core proposition (p. 14), often referred to as “the thesis of autonomy of formal grammar.” Such appearances are misleading. Chomsky’s discussions of autonomy have focused on a much narrower issue—namely, whether core semantic notions are required for the proper formulation of syntactic rules. His answer has been consistently negative, but this does not mean that semantic considerations are irrelevant to the formulation of syntactic analyses. By listing autonomy as a core proposition, Huck and Goldsmith imply that it is somehow intrinsic to the program. Yet, as Chomsky’s discussion of such hypotheses in his June 1974 lecture at the Linguistic Society of America’s Golden Anniversary Symposium (published as Chomsky 1975b) shows, a notion regarding the autonomy of formal grammar is simply an empirical issue to be resolved, not something on which the viability of the theory depends. It is also worth noting that this purported core proposition, which Huck and Goldsmith claim is carried over to the Aspects program (p. 26), was barely mentioned in the GS literature (with the exception of Lakoff’s [1972] argument concerning preposing of adverbs) and certainly was not a central part of the alleged debate between Chomsky and proponents of GS, as indicated by the fact that the discussion of the autonomy of formal grammar in Chomsky (1975b) contains no mention of arguments by generative semanticists other than that of Lakoff, which Chomsky judged as having no bearing on the autonomy thesis. Thus it is rather odd that a “core proposition” on which GS and Chomsky disagreed should not be the subject of extensive debate. It is also distinctly odd that Huck and Goldsmith do not tie the existence of deep structure to a core proposition of the Aspects program and the interpretivist research program that replaced it—if presumably this was a key issue around which the debates raged.
Whether there actually were any “debates” between Chomsky and the proponents of GS is controversial. Chomsky has denied he took part in such debates, with the exception of a paper given at a 1969 Texas conference (Chomsky 1972a; see Barsky 1997:151). In chapter 3, “Rhetorical Strategies and Linguistic Argumentation,” Huck and Goldsmith cite three case studies. The first involves McCawley’s complicated analysis of respectively constructions (McCawley 1968). Chomsky commented on the manuscript McCawley sent him in a letter dated 20 December 1967, to which McCawley replied on 17 January 1968. Then, on 10 December 1968, McCawley wrote again to Chomsky, complaining that Chomsky had misrepresented his argument in his class lectures. McCawley writes again to Chomsky on 8 January 1969 after seeing a draft of the paper eventually published as Chomsky (1971) which contains a critique of McCawley’s analysis. On 12 February 1969, Chomsky responds, saying essentially that he is not going to revise his manuscript because he does not feel that he has misrepresented McCawley’s argument. In his review of Chomsky (1972b), McCawley (1975) admits to bungling his argument concerning respectively and tries to restate the argument more clearly. Hardly a debate, even a private one. The next example is weaker still. Lakoff (1972) contains an argument against deep structure based on the analysis of adverbs. Chomsky (1975b) comments on this two years later in the Linguistic Society of America Golden Anniversary Symposium paper mentioned above. The third example concerns Postal’s 1969 paper on the verb remind (Postal 1970). Chomsky comments on this analysis in his 1969 Texas conference paper (Chomsky 1972a). The manuscript generates an exchange of letters with Ross and with Postal.
If the purpose of these case studies is to paint a portrait of Chomsky as an ideologue who defends his pet ideas (e.g., deep structure) by overwhelming his critics with powerful rhetoric, then Huck and Goldsmith are surely on the wrong track. Chomsky has always maintained that the existence of deep structure is an empirical issue. If deep structure is part of Chomsky’s linguistic ideology, then how do we explain his willingness to give it up in recent work (e.g., in Chomsky [1993], which Huck and Goldsmith cite in another context, but not to point out this highly relevant fact)?
Turning to the published literature, we again find little evidence to support the view that Chomsky was engaged in an intense debate with proponents of GS. Huck and Goldsmith cite Chomsky’s (1970) “Remarks on Nominalizations” (written in 1967) as “what was perceived to be Chomsky’s first published response to GS” (p. 28). This perception was off the mark, however, because that paper was written in response to the complicated transformational analysis in Robert Lees’s (1960) dissertation on nominalizations. Lakoff’s (1965) dissertation, which it has been claimed the paper was intended to refute, is only mentioned once in a couple of brief footnotes. Chomsky (1971; written in 1968) is another paper that might be interpreted as an attack on GS. In fact, it does contain critiques of two arguments against deep structure, McCawley’s (1968) respectively analysis and Lakoff’s (1968) analysis of instrumental adverbs. In addition, it contains a general argument against “any variety of semantically-based grammar (what is sometimes called ‘generative semantics’)” (Chomsky 1971:198). However, the main thrust of the paper is to reevaluate “the relation of syntactic structure to semantic representation in generative grammar” (Chomsky 1971:183). Chomsky’s concern here is to develop a theory of grammar that departs from a model where underlying structure is the sole interface for semantic interpretation—a characterization that applies equally to Aspects and GS. “Specifically, these modifications have to do with some possible contributions of surface structure to delimiting the meaning of a linguistic expression” (Chomsky 1971:183). The discussion of McCawley’s and Lakoff’s arguments, which he characterizes as “proposals of a more ‘semantically-based’ grammar that have been developed in important work of the past few years” (Chomsky 1971:183), arise quite naturally in the context of the paper, which attempts to sort out the contributions of underlying and surface structures to semantic interpretation. Note that almost from the outset, Chomsky was aware that surface structure played a role in semantic interpretation (see Chomsky [1965:224 n. 9], which cites examples in which the order of quantifiers at surface structure affects semantic interpretation, examples that were also mentioned in Chomsky [1957:100–101]).
Although Huck and Goldsmith’s evidence for a debate on deep structure between Chomsky and proponents of GS is unconvincing, it does give a clear impression that the two sides appear to have been talking past one another. In retrospect, it seems obvious that this would happen. Chomsky was committed to the psychological interpretation of grammar as outlined in the first chapter of Aspects (written in 1958–59). Chomsky comments in the introduction:
In LSLT the “psychological analogue” to the methodological problem of constructing a linguistic theory is not discussed, but it lay in the immediate background of my own thinking. To raise this issue seemed to me, at the time, too audacious. It was, however, raised in the review article [of Syntactic Structures] by Lees [1957]…. In rewriting LSLT for publication in 1958–9 I presented this alternative psychological interpretation as the framework for the entire study, and in later work have continued to do so. [1975a:35–36]
Thus, from the late 1950s on, Chomsky was increasingly concerned with the central problem of language acquisition—of how an individual acquires rich grammatical knowledge on the basis of impoverished information. His tentative solution was that a child must be born with sufficient mental linguistic structure. It did not seem that the analyses of GS and the powerful descriptive devices they employed would lead to the kind of theory that could address the problem of language acquisition. Note also that this was probably Chomsky’s motivation for attempting to provide, seven years after the fact, an alternative to Lees’s (1960) transformational analysis of nominalizations. The rich transformational apparatus Lees employed was not compatible with the goal Chomsky had set for linguistic theory. GS, on the other hand, was primarily concerned with constructing a theory that could account for semantic interpretation broadly construed (including language use). Moreover, GS, for the most part, rejected psychological interpretations of grammar.
Having disposed of the apparent myth of the deep structure debates between Chomsky and the proponents of GS, let us turn now to the purported paradox that motivates the title of this study. Here, too, there is a distinct lack of evidence. For there to exist a paradox of the sort proposed by Huck and Goldsmith, it should be the case that some linguist has claimed that GS was disconfirmed and that that linguist has also proposed a theory that adheres to the central claims of GS. Clearly, Chomsky does not fit the bill since he has never claimed that GS was disconfirmed. Nor are Huck and Goldsmith helpful in identifying any other linguist to whom the paradox applies.
This still leaves open the question of whether the central claims of GS have been incorporated into current work on generative grammar. Huck and Goldsmith identify three core propositions supposedly unique to GS:
Syntactic and semantic representations are of the same formal nature, i.e., they are labeled trees.
Semantic representations resemble the constructions of formal logic, with quantifier scope and anaphor-antecedent relations explicitly expressed.
Syntactic and semantic representations are related via transformations, [p. 20]
The first and second propositions given here are identical to propositions that Huck and Goldsmith give in characterizing the “core” of the LSLT program (p. 14). (Newmeyer [1996] points out that Lakoff and McCawley rejected the first proposition a few years after the advent of GS and that the second proposition was rejected when global rules were introduced.) So, if we eliminate the autonomy thesis from the list of core propositions, GS circa 1967 appears to be an extension of the LSLT program. Note that Chomsky’s assessment of the situation in 1969 was that the differences between GS and the Extended Standard Theory (EST) he was developing were not as radical as had been claimed:
There is an appearance of a considerable diversity of points of view—and to some extent the appearance is correct. However, I think that the dust is beginning to settle, and that it is now perhaps possible to identify a number of real, empirically significant theoretical questions that have been raised, if not settled, in this work. I also think much of the apparent controversy is notational and terminological—including many issues that appear to be fundamental and have been the subject of heated, even acrimonious dispute.
[Chomsky 1972a:63–64]
Chomsky sharpens this assessment by stating “…when some vagueness and faulty formulations are eliminated, I think we can show that generative semantics converges with EST in most respects” (Chomsky 1972a:74). The question to be raised about Huck and Goldsmith’s three core propositions of GS is whether they were disputed by Chomsky and other linguists working within the EST framework. If not, then the fact that contemporary syntactic analysis is compatible with them tells us nothing about ideology in linguistic theory.
Even a cursory reading of Chomsky (1972a) reveals that these alleged core propositions were not the subject of dispute between GS and EST. For example, consider the first and second propositions in light of the following:
More generally virtually any proposal that has been made concerning semantic representation can, in equally uninteresting ways, be reformulated so as to use phrase markers for semantic representation. In particular, this is surely true of Katz’s [1967] semantic representations. It is difficult to imagine any coherent characterization of semantic content that cannot be translated into some “canonical notation” modeled on familiar logics, which can in turn be represented with properly bracketed expressions where the brackets are labeled (phrase-markers). The whole question is a pseudo-issue. [Chomsky 1972a:75]
Given that Chomsky was willing to grant that semantic representations could be given as annotated phrase markers resembling constructions of formal logic and that it was uncontroversial that phrase markers could be related by transformation (virtually the definition of transformational rule), then the third proposition would pretty much follow as a consequence. (See, for example, Chomsky’s [1972a:182] analysis of quantifiers as well as the more developed analyses in Chomsky [1976], which are clearly precursors of May’s [1977] quantifier raising analysis.) Thus, these core propositions are not incompatible with EST, in particular Huck and Goldsmith’s characterization of the core propositions of EST given on page 43.
Of course, Chomsky did not accept the specific analyses of GS that attempted to spell out these core propositions, but his objections were based on empirical considerations, not on a rejection of basic assumptions. If according to Huck and Goldsmith’s account of the core propositions of GS there is no difference between GS and EST, then there should be no principled reason why GS analyses should not show up in contemporary work. However, it is far from obvious that specific GS analyses are incorporated into contemporary syntactic theory. Although Huck and Goldsmith claim at the outset that GS analyses are adopted in Chomsky (1981) and (1986), no specific examples are given. Instead, they offer some general comments about Chomsky’s proposals for a level of Logical Form (LF), which they note conforms to the core propositions of GS (pp. 44–45)—unsurprisingly, given the previous quote from Chomsky (1972a). The only specific example cited in ILT is May’s (1977) analysis of quantifier raising, which is claimed to be conceptually similar to the GS analysis of quantifier lowering. What Huck and Goldsmith fail to point out is that the GS analysis, unlike May’s, involves the claim that quantifiers are treated as higher predicates that must be lowered into their surface structure positions. So the claim is simply false at the level of details. Moreover, GS analyses are based on the fundamental assumption that underlying structures are semantic representations, an assumption that was rejected in EST and the work that developed from it. In EST and its decendants, underlying representations are not LF representations, and the latter are not semantic representations. At least since Chomsky (1975c), it has been assumed that LF representations are mapped onto meanings by additional semantic rules “interacting with other cognitive structures” (Chomsky 1975c:105), giving fuller representations of meaning. Given that current work makes almost no use of “semantic representations” as conceived under GS, it is difficult to maintain that such work has incorporated GS analyses of any sort.
The main point of ILT is to show how linguistic theory is ideologically motivated. Since Huck and Goldsmith’s case collapses so completely primarily because their interpretation of history turns out to be so at variance with the facts, one can only wonder whether their charge is more applicable to their own revisionist history than to the history they seek to revise.
