The history of a science: unreliable narrators and how science moves on

It may seem that something remains to be explained in Chris Knight's question: Why did Chomsky's model of linguistics change?

Randy Allen Harris
9 May 2018


This photograph of Noam Chomsky is courtesy of John Ohala. All rights reserved.Chris Knight's 28 March openDemocracy article on the motivations behind Noam Chomsky's increasingly 'abstract' linguistic theories is a curious one, an answer to a question no one but him is asking, built on shaky presuppositions, and, in any case utterly unprovable.

Frederick Newmeyer, in his 15 April commentary, says that Knight is "completely wrong" about this. I'm a little more cautious on this front, inclined not to be so definitive as Newmeyer, and I don't imagine he has any more insight into Chomsky's motivations than the rest of us; that is, all of us who are not Chomsky. But I am with Newmeyer in finding Knight's case unconvincing.

Speculative psychobiography

Knight believes that, after a revelation of some kind, Noam Chomsky dedicated his career to an unusable form of linguistics, a theory so bizarre and inscrutable that it would prevent the American military from putting it into bloody practice in natural-language interfaces that would let soldiers kill by literally saying the word.

Maybe he's right. I wouldn't know. But Knight doesn't know either. He can't. His belief is based only upon a speculative psychobiographical story through which Knight splices two facts: Chomsky's linguistics is pretty unusable, growing more so over decades of iterations, and Chomsky loathes the American military.

Chomsky is also something of a technophobe, doesn't believe in jet lag, has a fondness for cable-knit sweaters, prefers his jeans rolled up, and is always surprised by surprise parties, no matter how frequently they happen or how obvious the occasion (he'll be 90 in December; head to Cambridge with your noisemakers; catch him off guard).

Maybe there are some dots to connect here, drawing a line to childhood trauma or adolescent disappointment. But maybe these are just some things that are true about Noam Chomsky, sort of like his linguistics is pretty unusable, and he loathes the American military.

The problem with Knight's claim to know Chomsky's mind is that there is only one source of evidence that could possibly bear on the motivations governing that mind, direct testimony from its owner, and that owner is not fully reliable. Such testimony exists. Chomsky's opinion of Knight's argument is clear. In a letter to the London Review of Books, Chomsky calls Knight's story (in its original form, in the book, Decoding Chomsky) "slanderous." So, he's not down with it. But that hardly settles the matter. Chomsky is regrettably equivocal, at best, about his own story.

Love at first sight

Let's see what we can make of Knight's narrative. It goes like this: Chomsky's early linguistics was a darling of Pentagon funding because it looked so beautifully adaptable for their death machines; Chomsky became worried that his theories might actually work for such things; so, he began making them so abstruse and convoluted that no one could possibly find a use for them. The story starts and ends well: The Pentagon did indeed lust after Chomsky's linguistics, throwing handfuls of money at it, and the current version of his linguistics is indeed not very useful for talking to jet planes, or translating Hungarian, or getting Alexa to lower the thermostat a few degrees. It's the middle part that is dicey, Chomsky's conscience-driven repudiation of practicality.

Chomsky's early theory looked tailor made for natural-language interfaces. Its inception coincided with the rise of those new-fangled thinking machines, digital computers, and Chomsky courted an impression of close compatibility. Computer scientists were smitten, love at first sight. "I must admit to taking a copy of Noam Chomsky's Syntactic Structures along with me on my honeymoon in 1961," computer pioneer Don Knuth has said. "Here was a marvelous thing: a mathematical theory of language in which I could use a computer programmer's intuition."

"I must admit to taking a copy of Noam Chomsky's Syntactic Structures along with me on my honeymoon in 1961"

That linguistics / Artificial Intelligence alliance, which also closely implicated psychology, began most obviously at a 1956 symposium in Cambridge, widely held to be the Bethlehem of the cognitive sciences. If one were to set up a nativity scene for the birth of the cognitive revolution, there would be four wise men around the manger: psychologist George Miller, and the two groundbreaking AI scientists, Allen Newell and Herbert A. Simon, and wunderkind Noam Chomsky.

The wunderkind gave a hugely influential paper, “Three Models for the Description of Language,” which, speaking in a quasi-mathematical idiom, appeared to say that his transformational theory was the only game in town when it came to modelling language. Newell and Simon presented their "Logic Theory Machine,” demonstrating intelligent activity from an artificial source. The two papers sang in harmony. Chomsky's featured a "device" that could be modified in various, increasingly powerful ways to model language, Newell and Simon's a "machine" that could make logical inferences. Chomsky's derivations and Newell and Simon's proofs were methodologically interchangeable. Simon, for one, saw the papers as twins. Here's how he describes the nativity matrix:

Historically the modern theory of transformational linguistics and the information-processing theory of cognition were born in the same matrix – the matrix of ideas produced by the development of the modern digital computer.

Miller (whose classic 7±2 paper completed the trifecta) was deeply impressed with the serious, firm-jawed young Chomsky, going on to work with him for over a decade. "Other linguists had said language has all the formal precision of mathematics," he later noted, "but Chomsky was the first linguist to make good on the claim. I think that was what excited all of us." Everyone felt that the convergence of devices, machines, grammars, minds, mathematics, and logic represented at the symposium was a new paradigm, with Chomsky's theory as the lynch pin, a theory of mental processes with startling computational affinities.

Chomsky the magnet

Chomsky thoroughly integrated himself with the emerging computational community. His first job, as Knight documents, was an appointment to work on computer translation of natural languages. He reports being very candid in his interview with lab director and military scientist, Jerome Wiesner, telling him that the whole project had "no intellectual interest and was also pointless." But Chomsky’s theories quickly came to define the activities of the lab. His first appearance in the MIT journal and de facto organ of the new field, Mechanical Translation, was by the project director of the lab's machine translation project, Victor Yngve, who said there were two basic problems to overcome with machine translation, and "N. Chomsky … [was] working on a theory of grammar that gives many new and powerful insights into" one of them, the fundamental structure of language. Mechanical Translation was soon bristling with references to his theory and citations to his work; the word transformation became ubiquitous.

Military money was all over MIT at the time (still is), and Chomsky was a magnet for it. As Knight shows, Chomsky, his colleagues, and his students were all heavily funded by the US military to underwrite computer translation and natural-language interfaces for machines of death.

In Knight's story, Chomsky was a naïf, a starry-eyed boy from Philadelphia, swept up in the military-industrial hub of MIT. But at some unknown point in the mid 1960s Chomsky had a crisis of conscience, as the implications of his military entanglements became inescapable and the US war machine escalated in Vietnam. This crisis resulted in a revelation with two outcomes. It drove Chomsky, on the one hand, to his tireless, life-long activism against all forms of American imperialism. On the other hand, it drove him to begin developing a string of increasingly unusable and fanciful linguistic theories.

Knight sees the one, along with many of us, as an act of courage and hugely admirable dedication. Knight sees the other, and he may be completely alone in this, as a kind of intellectual tragedy. The great activist sacrificed his linguistics, Knight suggests, to prevent his research from ever being culpable in the death of US military targets. Perhaps. Perhaps not. Perhaps one. Perhaps the other. Only Chomsky knows.

Perhaps. Perhaps not. Perhaps one. Perhaps the other. Only Chomsky knows.

Chomsky’s unreliability

Let's see what he says, you might think, and end the speculation. Here he is now:

Knight’s crucial charge … is that military funding influenced my scientific work. There is a very simple way to verify the charge: determine whether (and if so how) the work changed from the time I was a graduate student at Harvard with no military funding, to my early years at MIT, when its funding was quite generally military, to subsequent years when I received no military funding at all. Answer: not in the slightest relevant way – which is doubtless why Knight evades this test. Exactly the same is true of the other researchers in the same programme. End of story. And an end to the slanderous charges against all of us.

The letter goes into odd territory after this, Chomsky seeming to think that Knight accuses him of colluding with the military, which is the exact opposite of Knight's claim. Knight makes very clear, in his 24 April follow up, that his whole story rests on the premise of Chomsky "refusing to collude."

But, back to Chomsky's "test." Knight may in fact be evading it, since neither of his openDemocracy articles mentions Chomsky's letter (published last year) or addresses the documentary challenge. But maybe he just missed it, because evasion would be a surprising response given how easy it is to build the argument that Chomsky scolds him for avoiding.

As a graduate student (no military funding), Chomsky's work is incredibly abstruse. He wrote the massive, and quite alien, manuscript, Logical Structure of Linguistic Theory (LSLT), described by a contemporary linguist in these terms: "A few [of us] found it very difficult; most found it quite impossible. A few thought some of the points were possibly interesting; most simply had no idea as to how it might relate to what they knew as linguistics."

Syntactic Structures, on the other hand, is widely admired as one of the most elegant, lucid and powerful books on linguistics ever written. It was written after he joined the Mechanical Translation project and acknowledges support by "the U.S.A. Army (Signal Corps), the Air Force (Office of Scientific Research, Air Research and Development Command), and the Navy (Office of Naval Research)." Syntactic Structures attracted a great deal of interest indeed, anchoring the "Chomskyan revolution," and its author attracted scads more military funding, for himself, his students, and his colleagues. Within a decade, in Knight's story, Chomsky is overcome with horror at the possible military applications of his work, and what do we see in his work? More complexity begins to show up, and he starts arguing that is a good thing. Where he had been insisting that simplicity was the prime virtue in grammatical modelling, he now begins saying such things as "it is often a step forward … when linguistic theory becomes more complex."

Frankly, I don't see any correlation myself, and we wouldn't have to look far to find some countervailing evidence: Chomsky wrote very challenging technical articles alongside quite exquisite and graceful pieces over this period and beyond. But Knight does have material for an argument. The inconclusiveness of Chomsky's test for Knight's story is not the only problem, however.

"We encounter an embarrassment" with Chomsky, as the venerable research professor and intellectual historian, Margaret Boden, has noted: he "cannot be relied on to tell the truth." He is demonstrably dubious on easily documented facts, so there is not the slightest reason to believe he could be relied on about attitudes and motivations.

The clearest case of his dubiousness, though there is no shortage, might be Chomsky's sanctioned biography, Robert F. Barsky's Life in dissent. It is notoriously riddled with factual misalignments, even spelling errors, that originate directly with Noam Chomsky, its authorship seeming to be little more than copying and pasting from Chomsky's emails for Barsky, no fact checking allowed. But one particularly outrageous and easily discounted claim should suffice to make the point.

Chomsky's comments on the role of his teachers and mentors are noteworthy for their implications of incomprehension and neglect; scandalously so with Zellig Harris. Harris (no relation, by the way) was his supervisor throughout his three degrees at the University of Pennsylvania, but Chomsky told Barsky such things as "Harris never paid the slightest attention to" his work, and "he didn't know what I was doing.” In another biography, by Wolfgang Sperlich, he said much the same thing, and elsewhere he has been even more explicit. “Harris never looked at my [BA or MA thesis] on generative grammar,” he has said, and “it’s next to inconceivable … that Harris looked at my Ph.D. dissertation or LSLT." Let's be clear what these accusations amount to. If true, they suggest stunning professional negligence on Harris's part; if false, they approach libel.

Chomsky claims that his supervisor never looked at any of his theses – all of which Harris had to sign off that he had read and approved for Chomsky to get the related degrees (witness Harris' signatures on Chomsky's dissertation below). The B.A. and M.A. theses, further, were based on materials that Harris had given to his student to work on, and were in an area on which Harris continued to publish during that period. It is improbable in the extreme that he wouldn't be a little curious. As for Chomsky's dissertation and LSLT, they very prominently included Harris’s most famous linguistic innovation, the transformation. And Harris totally ignored them?


The signatory page of Chomsky’s dissertation, with Zellig Harris’s signature.

Let's pretend for the moment that Chomsky’s characterization of Harris doesn't run decidedly counter to normal expectations of professionalism, ethics, scholarly responsibility, and behavioral probability, and pretend further that we aren't required to believe Harris was an academic fraud, and just look at documented facts – such as what Chomsky said somewhat earlier on the same matter. “While working on LSLT," he writes in the preface to its publication, "I discussed all aspects of this material frequently and in great detail with Zellig Harris, whose influence is obvious throughout.” Barsky in fact even mentions, in another book (seeming not to register the inconsistency), that he saw letters from Harris indicating that "Harris recognized the importance of Chomsky’s work” – something one might suppose required reading that work. Or, we could check with Harris. In a 1957 paper— the publication of his Presidential address to the Linguistic Society of America, no less – he offers this affectionate endorsement of his recently graduated student, specifically mentioning the dissertation in which Chomsky says Harris could not conceivably have had the remotest interest:

From a time when this work was still at an early stage, Noam Chomsky has been carrying out partly related studies of transformations and their position in linguistic analysis: see his dissertation, Transformational analysis (University of Pennsylvania, 1955); and his paper Three models for the description of language, IRE Transactions on Information Theory, IT-2, No. 3, 113-24 (1956); now also his book Syntactic structures … My many conversations with Chomsky have sharpened the work presented here, in addition to being a great pleasure in themselves.

This prominent and generous endorsement was only one of many. Long before his Presidential address, for instance, Harris mailed two letters on the same day to draw the attention of influential linguists to "a student of mine, A. N. Chomsky." He wrote these letters at the very heart of his supposed neglect, in December 1950 (Chomsky's BA thesis was submitted in 1949, his MA, a reworking of the same materials, was submitted in 1951). One letter went to Bernard Bloch, editor of the most prestigious linguistics journal, then as now, Language. In it, Harris notes that he has sent Bloch some of Chomsky's work (probably the BA thesis), praising it as "rigorous" and "detailed" and linking it to his own work on Hebrew. "I wonder whether you think," Harris adds suggestively, there might be a "point in publishing it." Harris's other letter that day went to Charles Hockett, praising Chomsky in exactly the same terms and again seeking publication for his thesis research.

These efforts, let's not forget, are for undergraduate work, the first of the theses Harris allegedly ignored. And Harris was a big gun. Bloch and Hockett would have taken these requests with the utmost seriousness. No publications resulted directly from either of these letters, so far as I can tell, certainly not the Hebrew work, but Bloch did sit up and take notice. Over the next few years, Language published Chomsky's review of a book on Hebrew and then a major theoretical piece on Harris's transformational grammar, and Bloch was very important in the promotion of Syntactic Structures.  

Harris's advocacy for Chomsky – exactly what one would expect of a conscientious supervisor who recognizes a brilliant student when he sees one and tries to advance his career – is very, very difficult to square with an image of Zellig Harris as barely aware of Noam Chomsky's activities.What are Chomsky's motivations for smearing his supervisor? Perhaps Knight has a theory. Not me, and, as we have seen, there is no point asking Chomsky. He simply isn't reliable about such matters.

Knight’s question

So, then, what about Knight's original question? Why did Chomsky's model change, as it assuredly did, from the early, seemingly computationally tractable, heavily formalized, system of highly specific rules that set computer scientists' hearts aflutter, to a network of interacting principles that puts rules into the deep background and eventually eliminated the most precious rule of all, the one that Chomsky inherited from Zellig Harris and rejigged in a way that made the "Three models" argument and Syntactic Structures so compelling – the transformation?

If there is an explanation called for in the apparently decreasing usefulness of Chomsky's work over the decades, better reasons are probably to be found in the history of science, than in his personal antipathy to the military.

To start with a near universal in scientific change: failure creates instability and scientists move on. Chomsky's transformations didn't work. Their original usefulness was illusory. They never worked for Machine Translation or natural-language interfaces, something Chomsky says he knew all along (remember his reported comments to Wiesner). We could end Knight's story here, then. Chomsky's theories never were useful, so, if he wanted to stymie the military, why not just stick with them?

We could end the story here, then. His theories never were useful, so, if Chomsky wanted to stymie the military, why not stick with them?

But let's stay with the story a bit longer. Chomsky's work did change, after all, in a series of quite dramatic developments. The first few changes surprised the field, but such shifts eventually came to be expected from Chomsky. There are surely reasons for this in Chomsky's personal psychology, but they can also be seen in terms of a familiar history-of-science pattern. When theories and methods fail, science changes, and transformations failed.

‘Abstract Syntax’

Transformations failed not just in their usefulness for translating the documents of foreign powers or controlling missile trajectories, issues many linguists never cared about in the first place. They failed in ways linguists always care about, too, for building explanatory models of linguistic knowledge.

In the mid-1960s, at exactly the same time as Chomsky's supposed sudden horror at the military making use of his theories, many of his followers pushed transformations to do more and 'deeper' linguistic work than they had hitherto been tasked with, in a direction initially known as Abstract Syntax. Chomsky found this work uncongenial. In response to this line of research (or, just because he felt like it) Chomsky proposed reducing the role for transformations and his own work became "less abstract" as Newmeyer sees it.

Maybe this is where we should end the story, then: in the allegedly post-revelation period, just as Chomsky was becoming very vocal in his opposition to the Vietnam war, to American military imperialism, and to toadying intellectuals, his approach to linguistics became less abstract, not more abstract, which would appear to scuttle Knight's whole argument.

But abstractness is not really Knight's metric. As Newmeyer notes in his response to Knight, "Chomsky’s theories have evolved considerably over the past 60 years, but [they have undergone] little overall change in terms of their abstractness." That's true, but abstractness is a red herring in all of this – not one, I think, deliberately dragged across our path, but leading our noses in the wrong direction all the same.

Knight has imported the abstractness critique from various 'Cognitive Linguists' who have lined up against Chomsky, but it's not really what he means. He describes Chomsky's post-revelation trajectory as "so utterly abstract and other-worldly – so completely removed from any practical application – that no matter what insights [Chomsky] came up with, nothing could possibly be used to kill anyone."

Abstract is just one in a string of terms he tries out as synonyms for what he really means: useless.

Uselessness and revolutionary shifts

The real issue for Knight is not abstractness but practicality, and Newmeyer would probably not dispute that Chomsky's current theories appear less useful for computational purposes – for automatic translation or natural-language interfaces – than they once did. Chomsky is no longer hobnobbing so much with computer scientists, telling them how to model language. The heart of his framework is no longer a system of rules that generate derivations which resemble output from anything like the Logic Theory Machine.

Back to the history of science for a moment: many observers regard the Abstract Syntax movement, which morphed into Generative Semantics, as a kind of reductio ad absurdum of transformational grammar. Generative Semantics sputtered out after a few years of wild success (with significant strains of its research sponsoring 'Cognitive Linguistics'), and many linguists heard that sputtering as the death rattle of the transformation. The field fractured away from the transformational hegemony that loomed in the mid 1960s, and various schools of theorists went on to develop non-transformational (and most of them, in fact, more practical) approaches to language.

But Chomsky, in the extraordinary way that he has done so often, swam powerfully against the current until he changed the current. He modified his model in surprising ways, attracted new adherents, fought the old ones, then did it again, and hung on to transformations longer than most, scaling them back and generalizing the ones that were left until a single transformation remained. He cranked out other ingenious and highly influential procedures, contrivances, and utensils to accompany that lone transformation, quarterbacking at least two more 'revolutionary' shifts in linguistics.

All this is standard history of science in the Kuhnian mold, though it is rather remarkable to find so many models displacing each other with such rapidity, and even more remarkable to find one man behind a series of shifts. Usually 'revolutions' come from a contending camp that displaces the status quo. That's too mundane for Chomsky. He does it all himself.

Usually 'revolutions' come from a contending camp that displaces the status quo. That's too mundane for Chomsky. He does it all himself.

Many insights were gained along the Chomskyan way, many facts uncovered, much progress made. But eventually Chomsky let his last transformation go as well; indeed, the whole notion 'rule of grammar' seems to have left his framework. There is one 'operation,' Merge, that does virtually all the heavy lifting for Chomsky, not only ex-transformational duties, but ex-phrase-structure-rule duties, ex-lexical-rule duties, and so on. The title of a paper by Chomsky's close collaborator, Robert C. Berwick, sums up the current model: "All you need is Merge."

Chomsky and Berwick report enthusiastically on the fruits of this Minimalist trajectory: "It is no exaggeration to say that more has been learned about languages in the past twenty-five years than in the earlier millennia of serious inquiry into language." Others might dispute the level of exaggeration here, putting the results of the Chomskyan program against the phenomenal accomplishments of historical and anthropological and typological linguistics that had the misfortune to come before the last twenty-five years, but there is no disputing that Chomskyan linguistics has done well for itself.

Chomsky has certainly had opponents, and they have certainly shaped the field, both its intellectual content and its sociology, and the alternate Cognitive Linguistics paradigm probably takes up more linguistic oxygen these days than Chomskyan linguistics (usually sporting the term Biolinguistics), but as Samuel Hughes put it "Chomsky … remains the 800-pound gorilla in linguistics." When he changes his mind, he has little trouble persuading others to pursue his new course, often in conflict with previous disciples defending his old course.

As a final note, we might also remark that Chomsky and Berwick express considerable optimism about the practical applications of Chomsky's current program, musing that computational parsing algorithms based on Merge might pay good returns, and there is more talk of computers in their book (Berwick is a computer scientist) than has been seen in Chomsky's work for a long time. This optimism provides – ho hum – yet another stick in the spokes of Knight's story about Chomsky's create-a-useless-model motivations; though, again, the endpoint of that story is certainly right. No one is lining up to build Merge into automated translators or natural-language interfaces, so far as I am aware. The military is not throwing money at Merge. Chomsky has also shifted his recent attention more toward biological explanations for the evolutionary appearance of language, a field which has about as much bearing on natural-language interfaces as geology or Sicilian cuisine.

But the current stage of Chomskyan impracticality, like all its previous stages, is largely explicable in terms of the push and pull of scientific programs as various cooperative and antagonistic scientists try to make sense of their domain, with failures, false-starts, re-analyses and elaborations along the way – albeit with the rather unusual caveat that one guy, or one 800-pound gorilla, has done much of the pushing and pulling. No doubt Chomsky's psychology has played some role along the way, probably in recesses of his mind other than his antipathy to the military, but I have no confidence we will ever find out what that role is.

Had enough of ‘alternative facts’? openDemocracy is different Join the conversation: get our weekly email


We encourage anyone to comment, please consult the oD commenting guidelines if you have any questions.
Audio available Bookmark Check Language Close Comments Download Facebook Link Email Newsletter Newsletter Play Print Share Twitter Youtube Search Instagram WhatsApp yourData