Critique of Behaviorism and Linguistic Theories in Academia

In this interview, we talk to Noam Chomsky about the intellectual environment in which generative grammar emerged.

Download | Spotify | Apple Podcasts | Google Podcasts

References for Episode 23

Primary Sources

Bloomfield, Leonard (1933). Language. New York: Henry Holt and Co.

Carnap, Rudolf (1936). ‘Testability and meaning’, Philosophy of Science 3.4: 419–471.

Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton.

Chomsky, Noam (1959). Review of Verbal Behavior by B.F. Skinner. Language 35.1: 26–58.

Chomsky, Noam (1975). The Logical Structure of Linguistic Theory. Chicago: University of Chicago Press.

Chomsky, Noam (1979 [1949]). Morphophonemics of Modern Hebrew. New York: Garland.

Goodman, Nelson (1951). The Structure of Appearance. Cambridge, Mass.: Harvard University Press.

Harris, Zellig S. (1951). Methods in Structural Linguistics. Chicago: University of Chicago Press.

Harris, Zellig S. (1965). ‘Transformational theory’, Language 41.3: 363–401.

Hockett, Charles F. (1968). The State of the Art. The Hague: Mouton.

Lashley, Karl (1951). ‘The problem of serial order in behavior’, in Lloyd Jeffress (ed.), Cerebral Mechanisms in Behavior, 112–136. New York: Wiley.

Lenneberg, Eric (1967). Biological Foundations of Language. New York: Wiley.

Miller, George A. (1951). Language and Communication. New York: McGraw-Hill.

Quine, Willard Van Orman (1960). Word and Object. Cambridge, Mass.: MIT Press.

Shannon, Claude & Warren Weaver (1949), The Mathematical Theory of Communication, Urbana: University of Illinois Press. MPI PuRe: 1964 edition

Skinner, Burrhus Frederic (1948). Walden Two. New York: Macmillan.

Skinner, Burrhus Frederic (1957). Verbal Behavior. New York: Appleton Century Crofts.

Secondary Sources

Chomsky, Noam (2021), ‘Linguistics then and now: some personal reflections’, Annual Review of Linguistics 7: 1–11. DOI: 10.1146/annurev-linguistics-081720-111352 (open access)

Hiż, Henry and Pierre Swiggers (1990). ‘Bloomfield the logical positivist’, Semiotica 79.3–4: 257–270.

Matthews, Peter H. (1993). Grammatical Theory in the United States: From Bloomfield to Chomsky. Cambridge: Cambridge University Press.

Newmeyer, Frederick J. (1986 [1980]). Linguistic Theory in America: The first quarter-century of transformational generative grammar. London: Academic Press.

Transcript by Luca Dinu

JMc: Hi, I’m James McElvenny, and you’re listening to the History and Philosophy of the Language Sciences Podcast, online at hiphilangsci.net. [00:19] There you can find links and references to all the literature we discuss. [00:23] Our focus in this series now shifts to American linguistics in the middle of the 20^th century. [00:29] To help us get our bearings in this environment, we’re joined in this episode by none other than Noam Chomsky. [00:37] So, when you began studying linguistics in 1946, the dominant school in America was that of the Bloomfieldians. [00:47] They were committed to the psychological doctrines of behaviourism, which saw human language as a kind of learned habit, and to a highly empirical approach to analysing languages, which limited itself to describing surface patterns attested in corpora. [01:04] One of the great breakthroughs that you made in introducing generative grammar was to seek explanatorily adequate grammars that abstract away from surface details in order to capture the underlying principles of language as an endowment of the species Homo sapiens. [01:20] So how did you arrive at this position, at odds with the mainstream of linguistics at the time? [01:27]

NC: Well, first, to clarify the facts, I was actually at the University of Pennsylvania. [01:38] The leading figure there was Zellig Harris, a major figure in modern linguistics and a person who I was quite close to, followed him closely. [01:55] He himself was not particularly taken with behaviourism, and he didn’t really… He wasn’t interested in any of the psychological interpretations of what the language faculty is. [02:09] His approach was basically data-oriented, procedures of analysis for any corpus of data. [02:20] He worked out the most sophisticated and detailed procedures which you could, in principle, use to apply to any material to get some structural analysis of it. [02:32] And the question of what all of this meant psychologically just basically didn’t arise; he didn’t take it very seriously. [02:42] Now, Leonard Bloomfield, who was the leading figure in linguistics in that period, he himself was a dedicated behaviourist, at least in one side of his brain. [03:00] His view was that, as he put it, language is a matter of training and habit; if there’s anything new, it’s analogy, whatever… [03:09]. [03:10] And that was basically dogma. [03:13] Charles Hockett and other leading figures all accepted that. [03:19] Now, I myself ran into hardcore behaviourism a couple years later when I went to Harvard. [03:27] Harvard was dedicated to really, basically to Skinnerian behaviourism, a very rigid, narrow form of behaviourism. [03:40] I got to Harvard in 1951. [03:42] Skinner’s William James Lectures, which later came out as his book Verbal Behavior, they were circulating at the time, and that was doctrine for psychology, for philosophy. [04:01] W. V. Quine, the leading, the most influential philosopher, was completely committed to it. [04:09] His book Word and Object came out some years later, was strictly Skinnerian, and that was basically dogma. [04:19] That’s where I really ran into it. [04:21] And you ask… To me, it just seemed total absurdity, and if you simply look at the facts, the elementary facts, of language, then nothing like that is happening. [04:39] I also came to recognize even when I was an undergraduate at Penn that the procedural approach is, it seemed to me, seriously flawed. [04:50] Linguistics was described as what was called a taxonomic science. [04:57] It just organized data into categories, types, distribution of the elements, and produced a kind of taxonomy of the language based on some data. [05:11] That’s, both taxonomic linguistics and behaviourism struck me as sharp departures from anything that science has been doing for hundreds of years. I mean science seeks explanation and understanding, not arrangement of data, and it does not accept the behaviourist doctrine that you don’t look inside to see what’s causing the phenomena; you just look at the phenomena. [05:49] Both seem to me radically anti-scientific. [05:53] When I got to Harvard, I was lucky to have run quickly into a couple of fellow graduate students who were also sceptics. [06:05] We became close friends. [06:09] One was Morris Halle, who I worked with for the rest of my life. [06:11] The other was Eric Lenneberg, who years later went on and founded modern biology of language. [06:21] But even as young grad students in the early ’50s, we already just couldn’t see any sense to the prevailing doctrines. [06:31] Actually, there were cracks beginning to show in the orthodoxy, so one of the major figures in actually modern psychology, one of the founders of cognitive psychology, George Miller — we later became friends, worked together — he was running some — he was an experimental psychologist, and he did an experiment which people didn’t pay all that much attention to, but it knocked the props out of a lot of the work that was happening. [07:10] He was working on signal analysis (which was a common topic then coming out of the Second World War): How do you detect signals from noise? [07:20] Well, if you have a noisy environment, the probability of detecting the signal is such and such. [07:29] Now, if you think it through, the standard idea was, if you have a sentence, the first word comes along, you have a certain probability of detecting it. [07:41] Second word comes along, same probability detecting that. [07:45] Multiply the probabilities, you get the probability of getting the two of them right. [07:50] Well, what that means is, as you go through the sentence, the probability of getting the words correct, all the words correct, declines radically — sharply, in fact, exponentially. [08:02] Well, he ran the experiments — sort of worked, until you got to the last word. [08:08] When you got to the last word of the sentence, all of a sudden the probability shot up. [08:15] Well, what was happening was pretty obvious, but you couldn’t accept it. [08:19] What was happening is that people understood the sentence, and then figured out what those noises were early on. [08:29] By now, there’s fairly sophisticated work on that which shows that you can’t interpret the first few words of a sentence until you figure out what the sentence is. [08:40] All right, but that was completely contrary to anything that was happening, anything that was assumed. [08:47] There was also other work. [08:50] Karl Lashley, great neuroscientist, in 1951, published an article on serial order and behaviour where he showed that if you look at fairly complex actions like the gait of a horse, you know, galloping, or playing arpeggios on a piano, or anything that has any complexity, you find properties that just totally can’t be fitted into the framework of behaviourist doctrine. [09:29] And nobody paid attention to that article. [09:32] In fact, it was unknown. [09:33] He was a Harvard Medical School professor, famous neuroscientist, and the article had a lot of that recognition in the brain sciences, unknown in psychology. [09:49] I found out about it from an art historian, Meyer Schapiro, who suggested I look at it, brought it to George Miller’s attention. [09:58] So cracks were beginning to develop, but the whole edifice was based on sand, and as soon as you began looking at it, it all fell apart. [10:08] This did not have much effect, I should say. [10:12] The doctrines remained pretty rigid. [10:15] There are still replicas of them, but I think by late ’50s, early ’60s, it was, should have been clear that these are totally untenable notions. [10:29] Quine never agreed [10:32]; we had long discussions about this. [10:35] Anyhow, in the late ’40s, just to go back to my own personal experience, I was… [10:42] I mean as a young student, I was kind of a true believer: “This is what you’re taught. It’s got to be right. Important people,” you know. [10:48] I was also studying with Nelson Goodman, the philosopher, who was working on constructional systems and the concept of explanation, explanation and simplicity in constructional systems. [11:08] I was interested in that. [11:10] I had to write an undergraduate thesis, and Harris suggested to me that I do a structural analysis along procedural lines of Modern Hebrew, which is a language I pretty much knew, sort of knew, so I started doing it the way we were taught to do: get an informant, do fieldwork, you know, ask the right questions, do the taxonomic analysis. [11:43] But after a couple of weeks of this, it struck me as completely ridiculous. [11:49] First of all, I knew all the answers he was giving, and secondly, I didn’t care about them. [11:54] I didn’t care about the phonetics of the language; it just didn’t interest me. [11:59] So I sort of dropped it and decided to do what just seemed reasonable — write an explanatory theory of the language — and, following Goodman’s ideas, tried to pursue the simplest possible theory, well, as soon as you started… [12:18] – and that becomes an explanatory theory. [12:21] That’s what was later called a generative grammar, which is basically a theory of the language, and as soon as you did that, it turned out that, first of all, you start getting interesting results, and secondly, the elements that entered into the explanatory theory could not possibly be reached by procedures of analysis on the data — which is a familiar fact in the sciences, you know. [12:51] You know, I mean, if you’re from Silicon Valley, maybe you do it, but if you’re a scientist, you don’t just do organization of data and hope somehow a theory will come out of it. [13:02] It’s not the way it works. [13:04]

JMc: What did Zellig Harris think about this new work that you were doing? [13:10]

NC: He didn’t see any point to it. [13:14] We were friends, so he didn’t say much, but later he wrote about it. [13:21] If you go to 1965, he had a article in Language which you can find in which he simply dismissed this whole approach as kind of, you know, based on mistaken sociology, sociological ideas about competing theories; he said, “I’m [13:45] not interested in that, it’s nonsense.” [13:48] He attributed it to Cold War psychology of pitting theories against one another — which is science, you know. [13:57] So by that time, we were perfectly friendly, but often different dimensions. [14:03] I should say that he assumed, and it was generally assumed, that linguistics had really reached its terminal point with his book on methods of structural linguistics. [14:18] Actually, that was my introduction to the field when I was just getting interested in it as a young kid, 16-year-old kid. [14:27] I met Harris, and I was interested, and he gave me the manuscript of the book, and that’s how I learned linguistics, just reading it. [14:39] Actually, I was proofreading it for him. [14:41] But it was assumed then that that is essentially the terminal point: We have the procedures, we know how to apply them to data; from now on, it’s just a matter of applying it to different languages. [14:56] Now, there were actually major linguists who basically stopped working at that point and turned to data collection because you can just feed it into the procedures. [15:07] Our courses in the late ’40s in linguistics were not on linguistics; they were on analysis of discourse. [15:17] Efforts to extend the methods of linguistic analysis presumably finished [15:24] to broader topics like analysis of discourse. [15:30] Actually, Harris had a couple articles on this in Language in the early 1950s, so this was kind of often a… [15:46] You know, the field was esentially — there were no fundamental questions in the field. [15:51] Similarly, in the behaviourist psychology that I ran into at Harvard, there were fundamentally no serious questions left; we had the answers, but the… [16:03]

JMc: Do you think that that might have been connected to a sort of logical positivist conception of science that each science should be compartmentalized with its own narrowly defined problems and then together maybe they will create a whole picture of science? [16:18]

NC: It was a period of search for what was called unified science based on those ideas. [16:27] That was a very major topic, and especially in places like Cambridge in the 1950s. [16:33] Information theory fit into it, especially Warren Weaver’s interpretation. [16:40] If you look at Shannon and Weaver’s book, Shannon had the technical material, Weaver had an essay which, it was a — pretty good scientist himself — in which he essentially indicated you should be able to cover a whole, lots of domains of human intelligence, others, by just applying these probabilistic statistical measures. [17:06] And so that all sort of fit together. [17:11] Now, it’s kind of ironic about logical positivism, because the logical positivists themselves had by then abandoned it. [17:20] If you look at, say, Rudolf Carnap, by the mid-’30s, he was already writing critical papers, ‘Testability and meaning’ and so on, where he was departing from the orthodoxy. But it’s one of these cases of cultural lag, the doctrines that were being questioned by the founders were taking over other disciplines that were looking at the original materials. [17:55] So Bloomfield himself was very taken with logical positivism at a time when Carnap, who I knew, was actually talking about psychoanalysis, you know, which was anathema to the logical positivists. [chuckles] [18:14] There’s a lot of irony when you look at the actual history. [18:18]

JMc: So these movements, so we’ve spoken about behaviourism and information theory has also come up, and I guess, another sort of umbrella discipline that was around at the time is cybernetics, which interfaced closely with information theory. [18:37] So all of these, apart from the sort of scientific inadequacy of these disciplines as you established, did you have any objections also to the way that these disciplines marketed themselves, the way they sold themselves? [18:55] For example, Skinner was very clear about having a utopian project. [19:00] He wrote Walden Two, for example, where he had this vision of shaping people’s behaviour to create a better world, and cybernetics, of course, came out of studying human-machine hybrids in World War II, you know, [19:17] servomechanisms for aiming anti-aircraft guns. [19:22] So do your own theories, which look to what it is that makes the species Homo sapiens unique, and look at human creativity, do your own theories promote freedom in opposition to these technocratic scientific theories like behaviourism and cybernetics and information theory? [19:43]

NC: Well, cybernetics, at least Norbert Wiener’s version of it, didn’t have that property. [19:53] A lot of the marketing did, but not Wiener himself. [19:56] In fact, he was sharply critical of it. [20:00] Claude Shannon, who I also knew, also had no interest in any of these ideas. He was a mathematician working on theory of information. [20:12] Skinner did, you’re right, and what he was doing just struck me as humanly grotesque. [20:20] I wrote about it in later years but didn’t at that time. [20:25] So I basically just disregarded it as complete nonsense, but the atmos… Actually, Cambridge, MA, was really the centre of most of these ideas. [20:42] That’s where these things are being developed, and it was a very euphoric period. [20:47] The assumption, general assumption, was, “We’ve broken all the barriers.” [20:55] This is against the background of, a sociopolitical background which is interesting. [21:02] Pre-World War II, the United States was pretty much a scientific backwater. [21:09] If you wanted to study physics, engineering, philosophy, you went to Europe. [21:16] You wanted to be a writer, you went to France, you know. [21:20] There was American science, but it was kind of a fringe phenomenon. [21:24] In fact, when I got to MIT in the mid-’50s, one of my jobs was to teach graduate students how to fake their way through French and German reading exams, which was an anachronism. [21:41] Pre-World War II, that’s where the literature was. [21:44] Post-World War II, it was in English. [21:49] They finally abandoned the pretence. There was a sense that Europe was finished, US is taking over, we won the war — richest country — lot of advances in wartime technology which gave a sense of a great future ahead led by the United States and the global system in the sciences as well. [22:15] When you got to Crick and Watson 1953, you know, and a way of relating biology to chemistry, it looked as if the next frontier was taking mental phenomena regarded as mysticism and integrating them with the natural sciences by the means of information theory and behaviourism. [22:46] So there was tremendous euphoria about that. [22:49] Actually, as I said, it was all built on sand, very anti-scientific. [22:55] But it took some time and some intellectual struggles to get over this. [23:00] Cognitive science was just barely beginning to emerge in the ’50s. [23:06] George Miller, Jerry Bruner, a couple of grad students, those I mentioned, but not much else, and it was just barely beginning to… [23:17] The one academic connection was actually George Miller, who shifted… In 1950, he was a committed behaviourist. [23:27] You read his book on language and communication around 1950, it’s pretty strict Skinnerian, but as distinct from any of the others, he was open-minded and interested in thinking about new ideas and approaches, and he simply changed his mind and attitude. [23:48] In part, it was his own experiments, like I describe, in part just new things coming in. [23:54] By the mid-1950s, we were working jointly publishing papers together and so on, and that was kind of an opening in the academic world, and of course MIT was quite open. [24:08] MIT was just a science university. [24:10] They didn’t care if you had any credentials at all, which was lucky for me, because I really had no credentials, but they didn’t care as long as the work looked interesting. [24:21] So that developed, took off there. [24:26]

JMc: Okay. [24:27] So you don’t see any connection between the theoretical direction that you took and your broader political interests. [24:37]

NC: Well, I came out of a radical political background. [24:42] In fact, that’s how I met Harris in 1946, same political background. [24:51] We met through political connections. [24:53] Actually, I was very disappointed in college my first year, and I was thinking of dropping out. [25:01] It just was so boring. [25:03] It was kind of like extended high school, which I hated, and I met Harris, a very exciting person. [25:11] We had the same political interests. [25:14] He had all kinds of knowledge and understanding that I was very much interested in, and at that point, he suggested to me that I start taking his graduate courses. [25:25] Later, I realized he was basically trying to encourage me to get back into college, so I started taking his grad courses and then started taking graduate courses in other fields like Nelson Goodman in philosophy, math courses and so on. And they were pretty tolerant. [25:49] They let me take the courses though I had no background… [25:54] And so I never really had an undergraduate education, just a kind of an eclectic mixture of grad courses that I was picking up here and there, and that turned out to be very useful, but just luck, you know. [26:12]

JMc: Okay. [26:14] Great. [26:15] Well, thank you for answering those questions. [26:17] That was excellent. [26:18] And is that your dog I can hear in the background?

NC: Yeah, oh, yeah. [26:22] Two of them. [26:23]

JMc: Oh, two of them. [26:25] What are their names? [26:26]

NC: Gus and Philly. [26:27]

JMc: Gus and Philly. [26:28] Okay. [26:29]

NC: I hate to say the words; they’ll start running to the door. [26:33]

JMc: Yeah. [laughs] [26:34]

NC: That’s how behaviourism works. [26:38]