Language & Information Lab.
Pius ten Hacken
Since the emergence of Chomskyan linguistics, one of the questions that have been discussed again and again is whether or not it constitutes a scientific revolution. At an early stage the break with American structuralism was perceived as a revolution, e.g. by Thorne (1965) and Bach (1965). This perception is still widespread, cf. Newmeyer (1986), although by no means unchallenged. With few exceptions, the question is discussed against the background of Kuhn's (1970) description of what counts as a revolution. Some of the papers which argue against the use of the term revolution for the emergence of Chomskyan linguistics constitute first and foremost an attack on Kuhn's philosophy of science. This is notably the case with Percival (1976).
As is well-known, Kuhn's text can be and has been understood in a variety of ways. One problem is that it uses a number of well-known terms in, at least at the time of its original publication in 1962, novel senses. For the term paradigm the confusion arising from the novel and arguably inconsistent use by Kuhn is well-documented, cf. Masterman (1970:61-66). For the term revolution it is perhaps less obvious, but no less problematic. In this paper I intend to show that by turning to Kuhn's (1957) case study of the Copernican revolution as a more extensive illustration of how the concept of revolution should be interpreted, the label of Chomskyan revolution for the emergence of Chomskyan linguistics can be supported in ways that seem to have escaped attention in linguistic discussion so far.
One of the connotations associated with the term revolution is that it is an event related to a point in time or a relatively brief period. The Glorious Revolution in 1688 and the October Revolution in 1917, to name but two typical examples, took place in a matter of days. It is not surprising, then, that a scientific revolution is also often associated with a point in time, e.g. the execution of an experiment or the appearance of a book. In Popper's (1959) model of science, where the falsification of a theory by an experiment entails the rejection of the theory, a revolution can be linked to a point in time quite easily. This seems to be what Bach (1965) has in mind when referring to the Chomskyan revolution. In modern discussions too, both opponents of the analysis as a revolution, e.g. Kaldewaij (1986), and defenders of such an analysis, e.g. Newmeyer (1986), tend to identify the point of the revolution with the publication of Chomsky's Syntactic Structures in 1957.
In the case of the Copernican revolution, if a date in history had to be determined, the result would inevitably be 1543, the year of publication of Copernicus's De Revolutionibus Orbium Coelestium. Kuhn (1957) points out, however, that there are many reasons not to identify this date with the Copernican revolution. The universe described by Copernicus is in many respects closer to the Aristotelian or Ptolemaic universe inherited from antiquity than to the modern conception. Kuhn argues that the Copernican revolution finished only with the publication of Newton's Principia in 1686, so that the Copernican revolution is a process of about 150 years.
Another respect in which Kuhn (1957) relativizes the importance of 1543 concerns the novelty of the idea of a heliocentric rather than geocentric universe. The heliocentric universe was first devised by Aristarchos of Samos in the third century bc, cf. Heath (1913). The proposal as such was therefore not sufficient to bring about a revolution. As Kuhn points out, the possibility of a heliocentric universe was even actively discussed among the scholastics in Europe and Copernicus must have known of this discussion. The discussion, however, did not go any further than demonstrating that classical arguments against a hypothetical heliocentric universe were not logically compelling.
As I will show now, many of the problems raised by opponents of the term Chomskyan revolution are derived from an intuitive or pre-Kuhnian use of the term revolution. Conversely, a remarkably strong parallel can be observed between the Copernican revolution as analysed by Kuhn and the Chomskyan revolution in linguistics.
Kuhn (1957) distinguishes three main phases of the Copernican revolution, which he associates with Copernicus, Kepler, and Newton, respectively. The first phase started when Copernicus, in his De Revolutionibus, proposed to transfer to the sun certain astronomical functions until then commonly assigned to the earth. The reason for doing so was that certain observations of the position of the planets, notably their apparent regression, could be explained more elegantly. Many other aspects of ancient astronomy were not changed. Copernicus assumed that the universe was a sphere of finite dimensions with all stars equidistant from its centre. The courses of the planets were described as circles with epicycles as in Ptolemaic theory.
The parallel between De Revolutionibus and Syntactic Structures is strong enough to justify the identification of the Chomskyan revolution with 1957 on the same footing as the identification of the Copernican revolution with 1543. Chomsky (1957:48ff.) proposed to transfer the focus of theoretical discussion from a methodology for deriving a grammar from a corpus of utterances to a methodology for evaluating competing grammars by means of a corpus. In the same way as the heliocentric hypothesis, this was a relatively small change which turned out to have far-reaching consequences only later. Many other assumptions of the preceding theory were not affected at this point. The concepts of constituent structure, transformations, and kernel sentences existed in the theory of Zellig Harris in much the same way. The definition of a language as a set of sentences (Chomsky (1957:13)) with the suggestion that this set is probably infinite so that it "cannot be identified with any particular corpus of utterances obtained by the linguist in his field work" (1957:15) expresses the same attitude as Harris (1951:13) when he states that "the analysis of a particular corpus becomes of interest only if it is virtually identical with the analysis which would be obtained [...] from any other sufficiently large corpus of material taken in the same dialect."
In other respects, there are striking differences between the books of Copernicus and Chomsky. Whereas De Revolutionibus is a highly technical book full of calculations that only well-versed astronomers could read, Syntactic Structures is accessible to a much larger audience. Besides, whereas Chomsky was not even thirty when his book was published, Copernicus was seventy and dying when his appeared. This would be as if Chomsky had continued working on his The Logical Structure of Linguistic Theory of 1955 and published it as the first work in the new framework towards the end of his life.
The similarities at this stage make it possible to counter the often-heard argument that Chomskyan linguistics does not constitute a revolution because Chomsky took over so many concepts from the previously dominant school of linguistics. The differences clarify why Copernicus's ideas took so much longer to gain followers. As Kuhn (1957:185ff.) describes, Copernicus's colleagues received his work with a mixture of admiration and disbelief, with the admiration concentrated on the mathematical calculations and the disbelief on the heliocentric hypothesis. Insofar as they were convinced at all, it took them an important part of their active lives as researchers to accept the hypothesis, so that little time remained to contribute to the elaboration of the new framework. Chomsky, on the other hand, could elaborate and teach his new ideas himself, influencing and accelerating their development directly.
In astronomy, it is only with Kepler, a generation later, that we enter a new period, in which at least some astronomers learned the Copernican system as a starting point. These astronomers were not hampered by an Aristotelian or Ptolemaic background, so that they did not look automatically for correspondences with the old model in working with Copernican concepts. As a consequence, they were able to get rid of certain remainders of the Aristotelian view of the universe in the new system. At the same time, they shifted the set of data taken into account and generated new, unanswered questions.
An example of the first of these consequences is Kepler's replacing circular orbits with epicycles by simple elliptical orbits. Another is the abolition of the model of a spherical universe with a centre. Kepler also abandoned many of the traditional observations which by their inaccuracy made it impossible to arrive at a proper description of the course of the planets. He replaced them by Tycho Brahe's observations, the most reliable naked-eye observations available. An entirely new type of data was provided by Galileo's observations through a telescope. In a number of cases, the data gathered by even the most accurate naked-eye observations were too crude for the theories in the new framework. Only with the help of the telescope was it possible to observe the phases of Venus and to establish that stars are much smaller than expected from the extrapolation of naked-eye observations. The former confirmed a prediction of the new model, the latter solved a problem raised by the new model.
Among the new questions was that of the force driving planets in their orbits. Copernicus had followed Aristotle in the assumption that circular movement around the centre of the universe was natural, but without circular movement and without a centre such an explanation was no longer viable. The answer proposed by Kepler was that the planets are pushed around their orbits by rays of moving force, the anima motrix, which emanates from the sun (Kuhn (1957:214)). Although Kepler was later able to quantify the relationship between the speed of a planet and its distance from the sun, such a mystical, Neo-Platonist solution was never widely accepted.
In the years after the publication of Syntactic Structures, Chomsky and his followers also removed some old theoretical assumptions, opened a new source of data, and generated new questions. Among the old assumptions dropped now are kernel sentences, replaced by increased recursion in the rewriting rules. Instead of introducing phonological strings by rewriting rules (1957:32), Chomsky (1965:84) assumes a lexicon.
As we saw above, the data of linguistics were taken to be a set of sentences, of which a corpus is meant to be a sample. Chomsky (1957) specifies in various places that the set of grammatical sentences is assumed to be given, e.g. "Assuming the set of grammatical sentences of English to be given," (1957:18), "the sentences of language, which we have assumed were somehow given in advance." (1957:85), but he expresses his dissatisfaction with this assumption, which "is clearly too strong" (1957:103). There are only two remarks which seem to point to the use of grammaticality judgements. The first, "the sentences generated will have to be acceptable to the native speaker" (1957:49f.) is not about initial data gathering, but about testing grammars by considering the other sentences predicted to be possible, in addition to the ones given in advance on the basis of which the grammar had been written. It should be compared with Hockett's (1954:232) statement that "one must be able to generate any number of utterances in the language [...] which will pass the test of casual acceptance by a native speaker." The second, "It is undeniable that "intuition about linguistic form" is very useful to the investigator of linguistic form (i.e. grammar)" (1957:93f.), is so general that it only opens the theoretical possibility of using native speaker judgements without specifying which ones and how. If Chomsky had had grammaticality judgements as data in mind in 1957 it would be difficult to understand why he chose to present a less satisfactory solution of the data problem and explicitly admit its unsatisfactory nature. In response to explicit confrontation with this suggestion, Chomsky soon afterwards states that "The empirical data that I want to explain are the native speaker's intuitions" (quoted by Matthews (1993:201) from the proceedings of the 1958 Texas Linguistic Forum). What I intend to show here is that the problem of determining the data of linguistics was not automatically or immediately solved as part of the first step in the Chomskyan revolution, but only in a second phase, where problems raised by the new perspective were explored and confronted.
The most important new question raised at this stage was the learnability question, later called Plato's problem. It appears prominently in Chomsky's Aspects (1965), but the answer presented there, based on phrase structure rules and transformations, is hardly more satisfactory than Kepler's anima motrix. As opposed to Kepler, however, Chomsky explicitly admits the unsatisfactory nature of the explanatory force of his theory at this point: "Clearly, it would be utopian to expect to achieve explanatory adequacy on a large scale in the present state of linguistics." (1965:26).
The last phase of the Copernican revolution in Kuhn's (1957) analysis was the search for an explanation of the forces that moved the planets in their orbits. Kuhn concentrates on the competing theories put forward by Descartes and Newton. The latter, based on gravity as the attractive force between bodies, won out in the long run, but doubts remained, especially at the philosophical level. By now, Newton's theory of gravity has been given up in favour of Einstein's. Observers around 1700 would have had difficulty seeing that the Copernican revolution had ended with Newton. Some may still have preferred Descartes's theory; others may have been painfully aware of the imperfections of Newton's theory and waiting for decisive improvements.
In linguistics, the second phase was concluded much more quickly than in astronomy, due in part to the fact that the original instigator of the revolution was still alive to direct it. The third phase may have been concluded with Chomsky's Lectures on Government and Binding (1981), but it is difficult to tell. LGB does give a first overview of a possible though imperfect explanation, comparable in function to Newton's Principia in astronomy, or as Chomsky put it: "What seems to be particularly exciting about the present period in linguistic research is that we can begin to see the glimmerings of what such a theory [i.e. a theory of UG explaining how language acquisition is possible with limited data, PtH] might be like." (1981:4). The model of principles and parameters has gained a wide following but both theoretical and meta-linguistic, philosophical doubts remain. For some linguists these doubts are too serious to accept Chomsky's explanation. As I will show in section 4 below, this does not imply that the comparison with astronomy automatically breaks down.
If we accept the analysis of the Copernican revolution proposed by Kuhn (1957) and the parallel with the Chomskyan revolution as sketched here, we can also counter another argument occasionally put forward against the Chomskyan revolution, e.g. by Kaldewaij (1986). It is claimed that the theories of Chomsky in (1957) and in (1981) are too far apart to speak of a single revolution, that Chomsky (1957) is closer to its predecessors such as Harris (1951), and that it would therefore be arbitrary to impose a break in 1957. Matthews, while recognizing that Chomsky (1957) "stood Post-Bloomfieldian linguistics on its head" (1993:31) distinguishes a First Chomskyan school with Aspects and a Second Chomskyan school with LGB as the major reference works. Given the above discussion we can interpret the three phases in the revolution as moving from observational adequacy to descriptive adequacy and finally to explanatory adequacy.
One of the reasons for accepting or at least not neglecting Copernicus's theory at an early stage was the accuracy of Reinhold's Prutenic Tables, computed by Copernicus's new mathematical methods, cf. Kuhn (1957:187ff.). Observationally, these tables were fairly accurate, but they did not describe the underlying mechanism correctly. Such a description is exactly what Kepler's model with ellipses achieves, but it does not explain why the planets move. A certain degree of explanatory adequacy is only achieved by Newton in his Principia. Wasow (1985) outlines a similar development in Chomskyan linguistics, from observational adequacy in Syntactic Structures via a focus on descriptive adequacy in Aspects to a more realistic claim to explanatory adequacy in LGB. Observational adequacy is clearly the focal point when Chomsky (1957:13) states that "The fundamental aim in the linguistic analysis of a language L is to separate the grammatical sequences which are the sentences of L from the ungrammatical sequences which are not sentences of L." In statements quoted in sections 2.2 and 2.3 above from Chomsky (1965:26) and (1981:4) we have seen that, even if in principle explanatory adequacy has always been the goal, the connection between actual work in linguistics and the discussion of explanatory adequacy has changed dramatically in the period between these two publications. If we therefore reject the unity of the development within Chomskyan linguistics we must also reject the unity of the Copernican revolution as concluded by Newton.
The most striking difference between the two scientific revolutions compared here is their duration. Whereas the Copernican revolution took several generations between its starting point and its conclusion, the Chomskyan revolution covers the time span of one person's working life. There are a number of factors that play a role here. One factor, mentioned already, is that Chomsky was a young man when he published Syntactic Structures. Another factor is that the linguistic community in the second half of the twentieth century is much bigger than the astronomic community in the sixteenth and seventeenth centuries. In addition, scientists publish their works at an earlier stage. Few researchers nowadays can afford to delay publication of their ideas as long as Copernicus, Kepler, and Newton did. These two factors of community size and publication habits combine to speed up the dissemination, testing, and elaboration of new ideas drastically. A fourth factor not to be neglected is Chomsky's remarkable intellectual versatility. Most scientists have considerably more difficulty in continuing to develop new ideas when this implies giving up other original ideas of their own. The fact that the three major phases of the Copernican revolution are linked to different researchers represents a much more typical situation. In fact, in linguistics too, many researchers who enter the Chomskyan research programme at one stage leave the mainstream after a certain number of innovations, holding on to a version of the theory which does not correspond to the latest developments (cf. Botha (1989:7-8)).
Finally, we have to be very cautious about the later stages or perhaps the conclusion of the Chomskyan revolution. Astronomers of around 1700 could hardly decide whether the Copernican revolution had come to an end, but with three centuries of hindsight we can. For linguists it is impossible to decide at this moment whether we are like the astronomers of 1700, who knew Newton's Principia, or rather like the astronomers of 1650. In 1650 Descartes had published his writings on the corpuscular universe and few if any people could foresee that his explanation of Kepler's descriptive model was not to last. For this reason I think it would be premature to discuss the impact of the minimalist programme in its relation to the Chomskyan revolution.
A point which, perhaps to the surprise of some readers, has not been mentioned here so far is the condition of consensus among scientists, usually associated with Kuhnian paradigms. When Kuhn (1970:10-22) introduces normal science as scientific research carried out within a paradigm, the general consensus among scientists in a field stands out as one of the main characteristics of the paradigmatic as opposed to the pre-paradigmatic period of a science. In applying this criterion to linguistics the fact that neither Chomskyan linguistics nor any competing school can claim general acceptance has induced some people to conclude that linguistics is still in the pre-paradigmatic stage (or that Kuhn's theory is wrong).
There are several reasons to reduce the absolute nature of the condition on consensus for paradigms even within Kuhn's model. Thus, consensus is one among various characteristics differentiating the paradigmatic from the pre-paradigmatic period. Other characteristics include the nature of data collection and the type of literature (Kuhn (1970:15, 20)). In a paradigm, there is a heuristics which guides data collection, and literature consists mainly of textbooks and technical articles not accessible to the general reader. In these respects Chomskyan linguistics has the same properties as Kuhnian paradigms. Furthermore, the consensus hypothesis runs into numerous empirical problems. In the history of optics, for example, the wave theory and the particle theory of light have long co-existed, with, in the paradigmatic period, the prevailing view alternating between them. In his 1969 Postscript, Kuhn prefers to refer to "the relative scarcity of competing schools" in a field of science in the paradigmatic period instead of a more absolute statement as to the consensus. Therefore, I assume that general consensus is not a necessary condition for a paradigm. Whether different paradigms follow each other, overlap with each other chronologically, or coexist alongside each other is far less important than how they influence the way in which research is conducted.
In conclusion we can say that the parallels between the Chomskyan revolution and the Copernican revolution as analysed by Kuhn (1957) are so striking that there is every reason to assume that Kuhn would have agreed to the use of the term Chomskyan revolution for the emergence of Chomskyan linguistics. Kuhn (1970) repeatedly takes the Copernican revolution as a prototype, so that any attempt to construe individual remarks in the latter book as arguments against such a use of the term Chomskyan revolution are most likely run counter to the intended interpretation of Kuhn's text.
Bach, Emmon (1965), 'Structural Linguistics and the Philosophy of Science', Diogenes 51:111-128.
Chomsky, Noam (1975 [1955]), The Logical Structure of Linguistic Theory, Plenum, New York & London.
Chomsky, Noam (1957), Syntactic Structures, Mouton, Den Haag, 1978 (13th impr.).
Chomsky, Noam (1965), Aspects of the Theory of Syntax, MIT Press, Cambridge (Mass.).
Chomsky, Noam (1981), Lectures on Government and Binding, Foris, Dordrecht, 3rd revised edition 1984.
Harris, Zellig S. (1951), Methods in Structural Linguistics, University of Chicago Press, Chicago, (repr. as Structural Linguistics, 1960).
Heath, Thomas (1913), Aristarchus of Samos: The Ancient Copernicus, Clarendon, Oxford, reprint Dover, New York, 1981.
Hockett, Charles F. (1954), 'Two Models of Grammatical Description', Word 10:210-231.
Kaldewaij, Jelle (1986), Structuralisme en transformationeel-generatieve grammatica, Foris, Dordrecht.
Kuhn, Thomas S. (1957), The Copernican Revolution: Planetary Astronomy in the Development of Western Thought, Harvard University Press, Cambridge (Mass.), repr. 1985.
Kuhn, Thomas S. (1970), The Structure of Scientific Revolutions, Second Edition, Enlarged, Chicago University Press, Chicago (orig. 1962).
Masterman, Margaret (1970), 'The Nature of a Paradigm', in Lakatos, Imre & Musgrave, Alan (eds.), Criticism and the Growth of Knowledge, Cambridge University Press, Cambridge. Page reference to the German translation Kritik und Erkenntnisfortschritt, Vieweg, Braunschweig, 1974.
Matthews, Peter H. (1993), Grammatical Theory in the United States from Bloomfield to Chomsky, Cambridge University Press, Cambridge.
Newmeyer, Frederick J. (1986), 'Has there been a 'Chomskyan Revolution' in linguistics ?', Language 62:1-18.
Percival, W. Keith (1976), 'The Applicability of Kuhn's Paradigms to the History of Linguistics', Language 52:285-294.
Popper, Karl R. (1959), The logic of scientific discovery, Routledge, London (repr. 1992).
Thorne, James Peter (1965), 'Review of Postal, Paul (1964), Constituent Structure: A Study of Contemporary Models of Syntactic Description, Mouton, The Hague', Journal of Linguistics 1:73-76.
Wasow, Thomas (1985), 'Postscript', in Sells, Peter (1985), Lectures on Contemporary Syntactic Theories; An Introduction to Government-Binding Theory, Generalized Phrase Structure Grammar, and Lexical-Functional Grammar, CSLI, Stanford, p. 193-205.
I would like to thank Fritz Newmeyer for useful comments on an earlier version. Of course only the author is responsible for the claims made in this paper.