Ideology and Linguistic Theory

Noam Chomsky and the Deep Structure Debates

John A. Goldsmith, Geoffrey J. Huck

• Publisher: Routledge
• Year: 2013
• ISBN:
  9781136159909
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Linguistics
• Index: Languages & Linguistics

In The Ideological Structure of Linguistic Theory Geoffrey J. Huck and John A. Goldsmith provide a revisionist account of the development of ideas about semantics in modern theories of language, focusing particularly on Chomsky's very public rift with the Generative Semanticists about the concept of Deep Structure.

1
Introduction

Many linguists today assume that theirs is an empirical and deductive science, and that scientific progress in the domain of their research is possible. They believe that they, like their colleagues in the physical or biological sciences, are able to contribute measurably to the growth of knowledge by formulating and testing a sequence of empirically more successful and encompassing theories — each improved theory arising upon the falsification of a previous one.¹

Consistent with this belief in the empirical and progressive character of theoretical linguistics is the received view of its recent history. Whatever the status of linguistic work prior to the emergence of the generative paradigm in the last half of this century, there is general agreement that today's generative theories are more successful in accounting for the facts than the generative theories of thirty or thirty-five years ago, and that one can point to instances in the intervening period when a particular theory has been falsified by comparison with a competing theory that was demonstrably superior to it.

While there may be no unanimity concerning which among the train of successful linguistic theories is in the forefront these days, just about everyone seems to agree that there was a significant theoretical failure during the last few decades of generative syntactic work. On just about everyone's account, the theory that failed was Generative Semantics. That Generative Semantics has become such an exemplary failure has, no doubt, something to do with its successes. It was a theory that for a time in the late 1960s attracted a large number of adherents, produced impressive results, and had a substantial and lasting impact on the course of future work in the field. The Generative Semantics movement also challenged, with a good deal of brio and flair, the alternative - known as Interpretive Semantics - that was then being developed by Noam Chomsky.

The oft-repeated story told about the great collapse of Generative Semantics when it came in the mid-1970s is fully compatible with the credo of progress referred to above: if the central claims of Generative Semantics were empirically disconfirmed - or perhaps worse, if they led to a theory of grammar which was so vague as to be uninteresting - then the rapid dissolution of Generative Semantics might be explained as the result of its having been rationally rejected by the linguistics community in favor of ultimately more successful and interesting alternatives.

This standard story is by and large the one that is told in Newrneyer's (1980, 1986) historical survey and that appears, in various guises, in popular texts (van Riemsdijk and Williams 1986), monographs (Jackendoff 1983), and research articles (Katz and Bever 1976).² But there is a difficulty with this account that a number of historians, including Newmeyer, have noticed - a difficulty that is growing more and more difficult to ignore. It is this: 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 (1986b), is similar enough in certain crucial respects to that painted by Generative Semanticists 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.

It might be maintained that contemporary theorists have simply recovered usable parts of a discredited approach that did not work because of the way it was put together and because certain other hypotheses that it relied upon were shown to be wrong. Such a view would have its merits: Interpretivists from the very start were keenly interested in the questions the Generative Semantics program raised and openly incorporated its results when this could be done in terms they could accept. In effect, they and their descendants have wanted to show that the semantic enterprise was one that they could run better. But this also suggests that if there were problems with Generative Semantics, those problems were not necessarily systemic. The challenge would then be to determine which isolable propositions of Generative Semantics would have required replacement for the theory to have survived the empirical tests that were undertaken. But no critique of Generative Semantics has ever been attempted along such lines.³

Indeed, looking back now on the debates of the 1960s and 1970s, one cannot help but be impressed by their inconclusiveness in case after case, by the way in which fragments of information were routinely transformed into facts which one side thought constituted crucial evidence but which the other side felt were misconstrued or required more study. Since more study rarely confirmed original hypotheses in exactly the form in which they had been offered, and since many of the ensuing arguments in any case sputtered off in tangential directions or were simply not seen through to a point where consensus could be reached, it might be more natural to ask whether anything at all was learned in the debates, or whether they simply demonstrate that linguistic theories up to now have been not so much scientific theories as untestable collections of a priori beliefs.

In this book we reject the standard story of the rise and fall of Generative Semantics and propose a different approach to understanding its advocates' disputes with Interpretive Semanticists. In general, we agree that, at least within each of the programs, there were empirical issues that were rationally pursued. In fact, we share the credo alluded to in the opening paragraph of this chapter that work undertaken during the generative period has led to a significant increase in linguistic knowledge, even if disagreements about particular analyses persist.

Nevertheless, we will conclude that the debates about deep structure failed to settle a number of the larger issues, although settlement is often presumed in current research. We will show that in many of these cases no metatheoretical conclusions could rationally have been drawn from the evidence offered, which was insufficient in itself to decide in favor of one or the other of the theories. At the same time, we will observe that Generative Semanticists and Interpretive Semanticists, while starting from conceptually different positions concerning the organization of grammar, were led to approach some important issues in quite analogous ways. The two programs were in fact quite complementary, and the tensions between them not only bound each to the other, but also steered them jointiy on to a more productive path than either of them individually might otherwise have taken.

In the following chapter we will review in some detail the chief arguments put forward by Interpretivists against the Generative Semantics program. We will first suggest a rough definition of empirical failure that we will use when we come to assess those arguments. Then we will briefly consider the development of both the Chomskyan program and the rival program launched by the Generative Semanticists in the intellectual contexts in which they arose. When we look at the Interpretivists' critique of Generative Semantics, we will do so primarily from the standpoint of the Generative Semanticists - which is to say that we will attempt to reconcile the criticisms with our reconstruction of what the Generative Semanticists were proposing. We will conclude that the arguments against Generative Semantics proposals were not strong, and that over a significant range of issues Generative Semantics approaches and Interpretive Semantics approaches were a good deal more congruent than their advocates on either side were at the time prepared to acknowledge.

Chapter 3 focuses on the rhetorical techniques used in the dispute. In particular, we will compare the logical content of some of the arguments with the force of the language in which those arguments were framed. In this way, we may better understand why each side found the other's approach so unpersuasive and, ultimately, so distasteful.

Chapter 4 considers why, if the Generative Semanticists did not accept Chomsky's arguments against their program, the movement fell apart as rapidly as it did in the 1970s. It will be suggested that Generative Semantics failed for reasons that did not have much to do with the quality of the results it had turned up.

We conclude in Chapter 5 with some thoughts about the meaning of the dispute about deep structure for contemporary linguistic work.

2
Gaps in the Paradigm

Mediational and distributional themes in theoretical syntax

Remarks on Development and Progress in Science

There are well-known problems with a simple progressive view of scientific history according to which knowledge grows when improved theories replace those which have been falsified in confrontation with the facts. As Kuhn (1970) and Lakatos (1970) among others have made clear, serious questions must be addressed before it can be maintained that scientific theories can be falsified at all. Since a theory embraced by members of an active research community is an ever-changing constellation of propositions, and since anomalies can often be accommodated by minor adjustments to the theory, the effect of a counterexample to a theoretical prediction need not be considered fatal to the theory itself. Moreover, as Duhem (1954 [1905]) brought into stark focus, theoretical propositions usually work in combination, so that when a prediction fails it is not always clear which proposition of the many comprised by the theory is at fault. Hence, a research community may, with good reason, reserve its concern when challenged by what the skeptics may call "failed crucial experiments," relying on the belief - sometimes quite correct - that future work will straighten things out. But then, if counterexamples have no force, the assertion that science progresses as falsified theories are replaced by improved ones is called directly into question.

Problems of this sort have of course caused some historiographers to give up hope that a coherent argument can be made for progress in science. The more reasonable view in the eyes of many is that not every element of a theory need actually be empirically justifiable for that theory to have, overall, an empirical character. A variety of models has been proposed in the philosophical, sociological, and historical literature which attempts to salvage a progressive science by appealing to just such a precept.

We will not be concerned here with the question of whether linguistics fits one or another of these models, although we will in a general way (but only in a general way) follow the scheme of Lakatos (1970). We take the position that linguistic research programs comprise a body of falsifiable propositions, but also that such programs have a sociological dimension as well. As to the former, we will assume that there are empirical propositions held by members of a research program which, if the facts warrant, the researchers are willing to revise or replace, and we will say that such propositions constitute the auxiliary hypotheses of the theory. But since predictions of a theory are generally based on a conjunction of propositions, when a prediction is falsified the choice of which particular proposition is to be discarded may remain in the hands of the researcher. When the researcher shows a consistent methodological policy of not letting certain propositions be damaged by disconfirmation, we conclude that he or she is treating those propositions as part of a protected core of the theory.

The propositions of the core in effect constitute strategies for theory construction.¹ Since they will not themselves be subjected to empirical test (although evidence in their favor may be occasionally collected and sometimes even vigorously sought), they may never be precisely formulated. But the coherence of the research program depends on general agreement by its members concerning the content of these propositions. And, to a significant extent, what these core propositions are is determined by what the members of the research program hold to be its ultimate goals. We will call any general statement which characterizes such a goal of a program an orientation, or an orientational proposition of the program. Like core propositions, program orientations are irrefutable in practice, but unlike those propositions, they express desiderata: "Our theory should ultimately explain ..." or "The goal of our enterprise is to explain ..." are forms such orientations commonly take.

The distinctions among auxiliary, core, and orientational propositions as drawn here are notional, but this is because the category of a proposition depends more on the context of its use than on its content. What one researcher may understand as a core proposition another may take to be an orientational proposition or an auxiliary hypothesis, and so on. Moreover, competing research programs can, and frequently do, share propositions at all levels, including orientations. However, when research programs have very different sets of orientations, their respective results may prove incommensurable.

In Lakatos's view, theories become successful, or are progressive, when continued adjustments of their auxiliary hypotheses lead to continued discovery of hitherto unexpected facts without the concomitant loss of empirical content to the theory. It follows from this that there might be any number of reasons why a research program would fail to generate new auxiliary hypotheses that are corroborated by the discovery of new facts. For example, it could be that crucial experiments keep turning up anomalies rather than corroboratory evidence; in such a case, if the program fails it would be reasonable to say that the theory has been empirically disconfirmed. But it may also be that the members of the research program have merely stopped doing constructive research, for lack of time, money, interest - or whatever.² Hence, it may logically be claimed that there can be empirical progress, even though not all failures are necessarily empirical failures.

Lakatos's particular strategy for evaluating progress has been criticized (see, e.g. Laudan 1977: 77 ff. and Suppe 1977: 664 ff.). Of course, anyone who wishes to argue that Generative Semantics was rationally rejected on the basis of its empirical failures must at least assume with Lakatos that the empirical contents of the theories of distinct research programs can be compared in some scientifically respectable fashion. Whether or not Lakatos's overall scheme should turn out to be defensible, we will suggest that it is not possible to locate clear empirical grounds in the arguments offered in the 1960s and 1970s on which a decision between Generative Semantics and Interpretive Semantics could be based. More generally, we see no reasonable measure of even the conceptual contributions of the two programs that would have selected one over the other, say, circa 1971. In subsequent chapters, we will suggest that linguists had at their disposal other, perhaps more salient grounds on which to make comparisons.

The Development of a Distributionalist Orientation For Linguistic Theory

Before discussing the particular issues which separated Generative Semantics and Interpretive Semantics, however, it will be useful to review briefly the context in which these theories arose. It must first be recognized that there are two distinct orientations with roots that go deep into this century under which research programs in theoretical syntax, broadly conceived, have been organized. The first orientation has assumed that linguistic analysis should be undertaken in an attempt to discover and explain the relationship between sound and meaning, between outer form and inner form. We will call this orientation the mediational orientation to emphasize that language on this view is something that mediates between two very different types of phenomena.

The second orientation has taken it that linguists' central task is to explain the patterning and distribution of the formal elements of language. An essential presumption of this orientation is that there is a formal structure to language and that it can be described and explained on its own terms. We will call this orientation the distributional orientation, emphasizing the importance of distributional features of language study on this view. The distinction between the mediational and distributional orientations will play an important role in what follows.

To be sure, these two orientations are not necessarily or logically incompatible, and they could jointly define a single program. Indeed, most linguists in this century have at one time or another explicitly embraced both. For example, the mediational relation between sound and meaning is crucial in the work of Leonard Bloomfield, although some have erroneously interpre...

Table of contents

• Cover
• Title
• Copyright
• Dedication
• Contents
• Preface
• 1 Introduction
• 2 Gaps in the paradigm: Mediational and distributional themes in theoretical syntax
• 3 Rhetorical strategies and linguistic argumentation: Three case studies
• 4 What happened to Generative Semantics?
• 5 Conclusion
• Appendix: Conversations with Ray Jackendoff, George Lakoff, John Robert Ross, and Paul M. Postal
• Notes
• References
• Index