The Behavior Analyst 1991, 14, 49-60 No. 1 (Spring)

Skinner and Chomsky 30 Years LaterOr: The Return of the Repressed*

Julie AndresenDuke University

If we have learned nothing else in thelast thirty years, we have learned to un-derstand the importance of such vari-ables as context and audience. In this par-ticular context and with this particularaudience, I thought it unlikely that any-one would come expecting me to resur-rect Skinner's work from its very deepgrave only to bury it again. Nevertheless,I felt that my rather bland title "Skinnerand Chomsky 30 Years Later" called fora subtitle to properly frame my histori-ographic account of the fate of Skinner'sfamous 1957 book Verbal Behavior andChomsky's more-famous 1959 review ofit that appeared in Language.

In searching for an appropriate subti-tle, I extended the idea of graves and ex-humations and hit on "The Ghost ofChristmas Past." The image of behav-iorism as ghost will resonate for those ofyou for whom the stimulus of the word"behaviorism" elicits a very strong neg-ative response by calling to mind thespectres of other words such as "manip-ulation" and "control" and by suggestingthe frightening reductions of human ac-tivity to that of "rats pressing bars." Iffor some ofyou behaviorism is a spectraldragon, your hero will certainly beChomsky who was, almost unilaterally,acknowledged to have slain that monsterwith the well-aimed lance of his 1959review.However, since my sympathies are

firmly with Skinner-and I make hasteto foreground this bias-I wished to

* The phrase "Return of the Repressed" in thiscontext is Seymour Papert's (1988, p. 9).

I wish to thank Barbara Herrnstein Smith, Stan-ley Fish and the students of the "Language andTheory" seminar at Duke University for their dis-cussions oflanguage through the Fall Semester 1989and, in particular, Stephen Home for having madeavailable the articles in the Daedalus issue on Ar-tificial Intelligence (Winter, 1988).

stimulate sweeter responses from the au-dience and, thus, hit on a second candi-date, namely: "Snow White Awakens."This subtitle plays on behaviorism as ourheroine and casts Chomsky into the roleeither of the huntsman sent to slay SnowWhite and bring back her heart as proofof the deed or of the wicked stepmotherwho forced down our heroine's throat thepoisoned apple ofessentialism and ques-tion-begging nativism. Of course, SnowWhite does not die; she merely falls intoa deep sleep. Recently, she has beenawakened by the kiss of Prince Charm-ing, a handsome connectionist modelknown as Parallel Distributed Process-ing. Connectionism, it might be said-allowing a mix of metaphors and fairystories-is "behaviorism in computer'sclothing."'

Again, however, since my romanticimpulses are generally under the controlof different intraverbal operants, I re-jected the Snow White allusion in favorof presenting my account as a psycho-drama with the subtitle: "The Return ofthe Repressed." It's a classic story of re-pression that began when strict behav-iorism transgressed the bounds ofdecen-cy by entering the realm of humanendeavor and the sacred circle of verbalbehavior. It found itself instantly drivenback by the might and deeply culturally-inscribed forces of rationalism, formal-ism, innatism, and universalism. Its re-turn is like all returns in that nothing everreturns unchanged: strict, old-fashionedbehaviorism has been updated by bio-logical thinking about the brain, most no-tably Gerald Edelman's work on what hecalls neural darwinism.For the period from the late 1950's to

'See Seymour Papert (1988, p. 9). Papert alsouses the Snow White conceit in his paper to tell thestory of the return of behaviorism/connectionism.

49

50 JULIE ANDRESEN

the early 1980's, I have identified 4 rea-sons for the behaviorism's repressionand/or the success of generative gram-mar: i) cognitive taste; ii) the legacy ofthe 1960's; iii) the power ofessentializinghumanism; and iv) the discipline of lin-guistics as it conceived of itself throughits textual tradition. Changes in thesesame 4 categories have provided a morepositive climate for behaviorism in thelate 1980's.

i) Cognitive TasteThere was a distinct aesthetic appeal

in the transformational approach. Else-where, I have referred to the Bauhaus"feel" ofgenerative grammar with its tastefor a powerfully abstracted Cartesiangeometric (Andresen, 1988). It was cooland modern, stripped of ornamentation,and it operated in the hushed world ofthe idealized speaker-listener in a stream-lined, dust-free speech community.Transformational derivations lookedcrisp and scientific on a page, and trans-formational analyses were valued for theirelegance and economy. Within theframework of transformational gram-mar, the study ofthe finished monologicutterance was an art.2By comparison (if anyone had both-

ered to read Skinner to compare it) Ver-bal Behavior, with its rush of details, itshumor and its eccentricities, must haveseemed cluttered and inelegant. There islittle having to do with language in thewidest possible sense that does not re-ceive Skinner's attention: "reading, writ-ing, word associations, translation, be-lief, wit, verbal 'slips' and distortions,Empsonian 'ambiguities,' verbal gamesand puzzles, hypnosis ... metaphor, ab-straction, grammar and syntax, logic (the

2 The phrase "finished monologic utterance" isfrom Volosinov who explains that: "guided by phil-ological need, linguistics has always taken as itspoint of departure the finished monologic utter-ance-the ancient written monument, consideringit the ultimate realium. All its methods and cate-gories were elaborated in its work on this kind ofdefunct, monologic utterance or, rather, on a seriesof such utterances constituting a corpus for lin-guistics by virtue ofcommon language alone" (1973[1929], p. 72, emphasis mine).

nature of assertion, negation and predi-cation), scientific method, and a finalchapter on no less than thinking" (Smith,1957, p. 4).Now, of all the criticisms levelled

against Skinner, no one at the time ac-cused him of"bad style." However, Ver-bal Behavior is a frankly noisy book, re-sounding with exclamations of Fire! andWater! and illustrated with a welter oflively examples designed, in a functionalanalysis, not to describe something's fea-tures but to show how something oper-ates. My point here is that Skinner's bookreads well in the late 1980's set againstthe postmodern aesthetic with its new ge-ometry. I refer, of course, to the fractalgeometry ofBenoit Mandelbrot who, op-posing himself to the rigors of Bauhausdesign, offers instead the model of thearchitecture of the Beaux-Arts, with itssculptures and gargoyles, its cornices andscrollwork. "Art that satisfies," Mandel-brot claims, "lacks scale, in the sense thatit contains important elements at all sizes.... A Beaux-Arts paragon like the ParisOpera has no scale because it has everyscale" (Gleick, 1987, p. 117). Verbal Be-havior satisfies similarly in that it, too,has every scale: from a major rethinkingof the conception of "language" itselfdown to the most "trivial" observations,e.g., Skinner's comment on the "delight"of a "good palindrome-for example, Aman, a plan, a canal-Panama" (1957,p. 292).And speaking of Mandelbrot, the new

science ofchaos has alerted us to the factthat the scientist can no longer ignore"very small influences." One thinks hereofthe supposedly "irrelevant" influencesas "memory limitations, distractions,shifts ofattention and interest, and errors(random or characteristic)" ofthe Aspectsmodel (Chomsky, 1965, p. 3). To put itbluntly, classical, rule-governed perfec-tion is out, and the Butterfly Effect is in,with all its local unpredictability, statis-tical variation and just plain static.3

3The Butterfly Effect is the name, half-jokingly,given to the phenomenon of "sensitive dependenceon initial conditions." In meteorology, for example,the "sensitive dependence on initial conditions"plays on the idea that "a butterfly stirring the airtoday in Peking can transform storm systems nextmonth in New York" (Gleick, 1987, p. 8).

SKINNER AND CHOMSKY 51

ii) The 1960's: Post-SputnikShock and the War in Vietnam

Here the topic is money, and the in-tersection is with algorithms. If I hadnever before been aware just how closelywere linked the funding successes ofTGGin the 1960's to the Artificial Intelligenceindustry at M.I.T., I became fully con-scious ofit in a recent article by SeymourPapert (1988). In this article, Papert ac-knowledges that he and Marvin Minskywere responsible for having killed in thedawn ofcybernetics the early attempts tobuild neural networks with features in-herited from the study ofthe brain -whatwould return two decades later as con-nectionism. Instead, Papert and Minskywere devoted to building models of in-telligence out ofcomputer programs, withtheir formalizable domains, computa-tional complexity and algorithmic pow-er. Papert states candidly that he wasdriven to argue vigorously against theearly neural nets "from the fact that fund-ing and research energy were being dis-sipated on what still appear to me ... tobe misleading attempts to use connec-tionist methods in practical applica-tions" (1988, pp. 4-5). The fact remainsthat, in the funding game, Papert andMinsky's computer programming ap-proach to intelligence triumphed overthose machine models with no "in-nards," that is with no innate dispositionto acquire particular behaviors.The successes of AI parallel the suc-

cesses ofTGG to capture NSF and NDAgrant monies, and none of the fundingcan be divorced from the war in Viet-nam. Frederick Newmeyer has noted that"the earliest money from the armed ser-vices filtered into the M.I.T. ResearchLaboratory of Electronics and variousmechanical translation projects." Trans-formational research was deemed bysome in the Air Force to hold promiseof programming 'command and control'computer systems ofuse in planning andexecuting military operations in the mid-60's (Newmeyer and Emonds, 1971, p.288).On a grander scale than even that, ar-

tificial intelligence and the transforma-tional approach benefitted from the

weight of the tradition of symbolic in-formation processing, which is the wayof all Western philosophy, and not justDescartes and his descendants. WhatPapert and Minsky and Chomsky tookfor granted was what they had inheritedfrom the tradition extending from Soc-rates to Kant, namely the idea that:

... understanding a domain consists in having atheory of that domain. A theory formulates the re-lationship among objective, context-free elements(simples, primitives, features, attributes, factors, datapoints, cues, etc.) in terms of abstract principles(covering laws, rules, programs, etc.). (Dreyfus andDreyfus, 1988, p. 25).

In one of Chomsky's recent books,Knowledge ofLanguage (1986), he is stillcommitted to making such statement as:"children unerringly use computationallycomplex structure-dependent rules ratherthan computationally simple rules . . .";and ". . . the language faculty appears tobe, at its core, a computational systemthat is rich and narrowly constrained instructure and rigid in its essential oper-ations, nothing at all like a complex ofdispositions or a system of habits andanalogies" (1986, pp. 7, 43, emphasismine). Like I said, transformational for-malisms looked good-even familiar-on a page.4

4 In his critique ofinformation processing modelsofhuman intelligence, Edelman writes: "Accordingto information processing models, neural signalsfrom the periphery are encoded in a variety ofwaysand are subsequently transformed by various nucleiand way stations; finally, they are retransformed ina variety ofways by increasingly sophisticated relaysystems culminating in cortical processing and out-put. Perforce, this view puts a very strong emphasison strict rules for the generation of precise wiringduring the development of the brain. Such modelsstrongly rely on neural coding (Bullock, 1967) andon the transfer of information from one particularneuron to another. This view also makes an as-sumption about the nature ofmemory which it con-siders to occur by representation of events throughrecording or replication of their informational de-tails. The notion of information processing tendsto put a strong emphasis on the ability ofthe centralnervous system to calculate the relevant invari-ances of a physical world. This view culminates indiscussions ofalgorithms and computations, on theassumption that the brain computes in an algorith-mic manner (Marr, 1982). Categories ofnatural ob-jects in the physical world are implicitly assumedto fall into defined classes or typologies that are

52 JULIE ANDRESEN

For the last thirty years at least, be-haviorism and innatism have shared in-terlocking and inverse fates. The new andremarkable features of the more recentneural nets is that no one programs them.They are "trained" to do their tasks bya rigorous behaviorist process ofexternalassociation of stimuli with reinforce-ments. However, since neural nets areusually simulations run on digital com-puters, it is sometimes doubted whetheror not neural nets offer something revo-lutionarily new in artificial intelligence.Yet, no one doubts that the excitementgenerated by "the return of the re-pressed" neural nets is genuine; and whileneural nets are not intended to be modelsof living organisms, they are inspired by

accessible to a program. Pushing the notion evenfurther, proponents ofcertain versions ofthis mod-el are disposed to consider that the rules and rep-resentations (Chomsky, 1980) that appear to emergein the realization ofsyntactical structures and high-er semantic functions of language arise from cor-responding structures at the neural level. If statis-tical variation enters at all into such a view of thebrain, it is considered in terms of noise in a signal,which in information processing models is taken tobe the main manifestations of variation" (1987, p.38).

Table 2.1Some Unresolved Structural and Functional Issues

in Neuroscience

ProfferedFacts Explanations

Precise, prespecified, point-to-point "Noise"wiring is excluded.

Uniquely specific connections cannot "Derived at higherexist. levels"

Divergent overlapping arbors imply "Codes"the existence of vast numbers ofunidentifiable inputs to a cell.

The majority of anatomic connec- "Silent synapses"tions are not functionally ex-pressed.

Major temporal fluctuations in maps; "Alternative sys-unique maps in each individual; tems"variability of maps in adults de-pendent upon available input.

Extensive generalization in object "Hidden cues"recognition without the need forlanguage.

Unitary appearance to the perceiver "Algorithms, com-of perceptual processes that are in putations, invar-fact based upon complex parallel iants"subprocesses.

the neurobiological modeling of Edel-man who has exposed the inadequaciesof the information processing model ofthe brain (1987, p. 38).5

iii) The Power ofEssentializingHumanism

If we want to find a reason why Skin-ner's approach to verbal behavior did not"take," we need look no farther than the"common-sense" theory oflanguage andcommunicative reasons for language usein our culture. The ability to talk is partof our humanity, and a central axiom inthe conception ofthat humanness is par-ticularly hard to abandon: namely, thatthe human being is an actor, rather thana locality (see Vargas, 1986, p. 129).Chomsky's 1959 review of Skinner

played up and played on the worst fearsengendered by behaviorist approaches tohuman activity. Chomsky's review is 31pages long. On 13 ofthose pages, Chom-sky refers to rats or Skinner's bar-press-ing experiments, often more than onceper page-although nowhere in VerbalBehavior is there mention of rats and, inmy first and recent reading of Verbal Be-havior, I found Skinner's book to be al-ways fresh, often amusing and sometimesidiosyncratic- in short, very human.Nevertheless, Chomsky framed his entire

5 In the New York Times BookReview ofSunday,December 24, 1989, George Johnson has written areview of Jeremy Campbell's new book The Im-probable Machine. What the Upheavals in ArtificialIntelligence Research Reveal About How the MindReally Works. Johnson is critical ofCampbell's po-sition that neural nets are truly new. Johnson de-fends Minsky et al. by writing that "no one in ar-tificial intelligence ever believed that the brainprocessed information in the same way as the I.B.M.tabulating your telephone bill.... For those tryingto understand intelligence, the computer was not ablueprint but an inspiration and a tool, the perfectdevice for modeling complex systems." Johnsondoes concede that: "True, a few people in artificialintelligence have been preoccupied with using sym-bolic logic to simulate the mind."Chomsky might be identified as having been pre-

occupied with symbolic logic, and he might also bementioned as one who did believe that the brainprocessed information like a computer.Johnson does not mention Edelman's neuro-

biological work in his review.

SKINNER AND CHOMSKY 53

review of Verbal Behavior with a five-page discussion of Skinner's Behavior ofOrganisms (1938) and recurrently asso-ciated Skinner's innovative terminologyfor human language-use with rats press-ing bars.6 The most egregious examplewas his dismissal ofthe term "verbal be-havior." Some 19 pages into the review,Chomsky writes:Consider first the term 'verbal behavior' itself. Thisis defined as 'behavior reinforced through the me-diation of other persons.' The definition is clearlytoo broad. It would include as 'verbal behavior, forexample, a rat pressing the bar in a Skinner box, achild brushing his teeth, a boxer retreating beforean opponent, and a mechanic repairing an auto-mobile. (1959, pp. 44-45, emphasis mine)

Chomsky's rhetorical move, that is, in-terlarding his review of Skinner with re-peated references to "rats pressing bars,"was extremely effective, but the scare tac-tic came at a price. With this excoriatingreview, Chomsky might be said to haveprogrammatically exiled pragmatics fromlanguage theory.Now, however, 30 years later, Skin-

6 The relevant passages from Chomsky's revieware:

"Skinner specifies 'response strength' as the basicdatum, the basic dependent variable in his func-tional analysis. In the bar-pressing experiment, re-sponse strength is defined in terms of rate of emis-sion during extinction. Skinner has argued that thisis 'the only datum that varies significantly and inthe expected direction under conditions which arerelevant to the "learning process."' In the bookunder review, response strength is defined as 'prob-ability ofemission"' (1959, p. 34, emphasis mine);

and: "Verbal operants are classified by Skinnerin terms of their 'functional' relation to discrimi-nated stimulus, reinforcement, and other verbal re-sponses. A mand is defined as a 'verbal operant inwhich the response is reinforced by a characteristicconsequence and is therefore under the functionalcontrol of relevant conditions of deprivation oraversive stimulation' (35). This is meant to includequestions, commands, etc. Each ofthe terms in thisdefinition raises a host of problems. A mand suchas Pass the salt is a class of responses. We cannottell by observing the form of a response whether itbelongs to this class (Skinner is very clear aboutthis), but only by identifying the controlling vari-ables. This is generally impossible. Deprivation isdefined in the bar-pressing experiment in terms oflengths oftime that the animal has not beenfed orpermitted to drink. In the present context, however,it is quite a mysterious notion" (1959, p. 45, em-phasis mine).

ner's original formulations and particu-larly the idea of"verbal behavior" do notseem so misguided or so threatening:In defining verbal behavior as behavior reinforcedthrough the mediation of other persons we do not,and cannot, specify any one form, mode or medi-um. Any movement capable of affecting anotherorganism may be verbal. We are likely to single outvocal behavior, not only because it is commonest,but because it has little effect upon the physicalenvironment and hence is almost necessarily ver-bal. But there are extensive written languages, signlanguages, and languages in which the "speaker"stimulates the skin of the "listener." Audible be-havior which is not vocal (for example, clappingthe hands for a servant, or blowing a bugle) andgestures are verbal, although they may not composean organized language. The skilled telegraphist be-haves verbally by moving his wrist. Some of theseforms normally arise only after vocal behavior hasbeen established, but this is not necessarily so. Writ-ing and typing may be either primordially verbalor transcriptions of a prior vocal form. Pointing towords is verbal-as, indeed, is all pointing, since itis effective only when it alters the behavior ofsome-one. (1957, p. 14)

This passage will not strike as odd thereader used to the arguments of JacquesDerrida, who has contributed to decon-structing and blurring the facile dichot-omies that have troubled Western phi-losophy for centuries, including that ofwritten versus spoken. Nor would thispassage strike as odd the reader familiarwith the work of J. L. Austin, who con-stantly integrates his analysis of perfor-mative utterances with illustrative per-formative actions of the type: "supposeI bow deeply before you; it might not beclear whether I am doing obeisance toyou or, say, stooping to observe the floraor to ease my indigestion." Austin ex-plains the case as one of "actions whichare non-linguistic but similar to perfor-mative utterances in that they are the per-formance of a conventional action (hereritual or ceremonial)" (1975 [1962], p.69). The difference between Skinner andAustin on this score is only that for Skin-ner the first act (of ceremonial bowing)would count as verbal, but non-vocal.What does strike the contemporary

reader in Skinner's description of"verbalbehavior" is the radical non-automonyof that description. Not only does Skin-ner fail (or refuse) to separate the writtenfrom the spoken-a major disciplinary

54 JULIE ANDRESEN

heresy 30 years ago -and the vocal fromthe gestural, he also fails (or refuses) toseparate out an "autonomous speakingagent." Nor does he separate the speakerfrom the listener in any given speech ep-isode (see p. 84ff), including those in-stances when the speaker behaves as hisown listener as in "thinking."And, to be sure, Skinner does not sep-

arate human behaviors from the behav-iors of other creatures. The memorablefirst paragraph of Verbal Behavior pro-poses a double continuity of verbal be-havior with all other human behaviors,and of verbal behavior with the behav-iors of other creatures:Men act upon the world, and change it, and arechanged in turn by the consequences oftheir action.Certain processes, which the human organism shareswith other species, alter behavior so that it achievesa safer and more useful interchange with a partic-ular environment. When appropriate behavior hasbeen established, its consequences work throughsimilar processes to keep it in force. If by chancethe environment changes, old forms of behaviordisappear, while new consequences build new forms.(1957, p. 1)

Here is one of Skinner's enduring in-sights, then, that of "selection by con-sequences," where behaviors, both for thespecies and for the individual, are se-lected and maintained by effects in a Dar-winian framework (see also Skinner,1981). Skinner has insisted "that no rep-utable student of animal behavior hasever taken the position that the animalcomes to the laboratory as a virtual ta-bula rasa, that species' differences are in-significant, and that all responses areabout equally conditionable to all stim-uli" (1966, p. 1205). And this in spite ofthe routine charge that behaviorist mod-els assume that a general theory oflearn-ing could hold across species. Current re-search certainly emphasizes the fact thatparticular evolutionary features condi-tioned by niche influence the learning andbehavior ofindividual species (Edelman,1987, p. 293). In Neural Darwinism, forexample, Edelman acknowledges his re-lationship to a (Skinnerian) type of op-erant conditioning.7 He is, however, crit-

7 In the following passages, Edelman summarizesthe relation between classical and operating con-

ical of the (Skinnerian) type of radicalbehaviorism only insofar as it is envi-ronment-driven and allied to a covert no-tion of an empty animal driven by cat-egories from the outside. Edelman insiststhat brain structures "themselves arepop-ulations upon which selection acts ...[and] that such selection is the basis ofperceptual categorization" (1987, p. 12).

It is important here to emphasize that,for Edelman, the world is initially an un-labelled place, that it does not have anyparticular information in it to instruct the

ditioning. At the end, he identifies his own positionas not representing an extraordinary divergence froma synthesis of the two types of behaviorisms:"Leaming is a specific change in the animal re-

lated to a positive or negative outcome, using anevent or its absence as a signal for something else.Classical conditioning involves reliable presenta-tion ofa neutral conditioned stimulus (CS) prior toa hedonic stimulus (the unconditional stimulus, orUS); the CS is a predictor or signal for the US, andthe animal reacts to the CS as if it is in anticipationofthe US. In contrast to this classical, or Pavlovian,conditioning, operant conditioning involves twophases. The first is behavior that leads sooner orlater to reward or punishment, a kind of control byconsequences (Skinner, 1981). The second phaseconsists of recurrence of adaptive behavior whenthe animal is again presented with the original sit-uation."These two forms of conditioning are closely re-

lated: the selection of the behavior in a Pavlovianresponse depends upon the animal's ability to pre-dict on the basis of the relation between the par-ticular CS and the US; in the operant situation, thediscriminative stimulus consists of all those envi-ronmental features (leading to categorization) thatcan control behavior. In both cases ofconditioning,a change in expectancy, signaling a change in in-temal state based on categorization, is required."A way ofsummarizing the difference in the two

modes is to say (Staddon, 1983) that classical con-ditioning is open-loop (or a procedure for assigningvalue to a neutral stimulus); in contrast, operantconditioning is closed-loop and leads to a changein the priority of action. While in this latter caseaction is part ofthe animal's categorical represen-tation, classical procedures relating contingenciesare nonetheless required to reinforce operant be-havior. It is the Pavlovian mode that allows ananimal to define a situation according to valuesdepending upon prior innate and previously formedassociations. In the operant mode, the animal mustnot only select candidate stimuli related to valuebut also select stimuli to allocate behavior. Givenour view ofglobal mappings and ofgestures as partof categories, this does not represent an extraor-dinary shift in interpretive point of view" (1987,pp. 297-298).

SKINNER AND CHOMSKY 55

brain like a computer tape, nor then doesthe brain act as a computer to interpretthat information. Thus, the question isnot whether human verbal behavior isbiological but rather at what level ofspec-ificity. For Edelman, it is deeply biolog-ical that the brain is a general-purposeperceiving and thinking organ not inher-ently marked for a prescnrbed function ormeaning. It is, for Edelman, deeply bi-ological that the brain has no built-inprogram, that individual groups of neu-rons do not have any a priori fixed pur-pose, and that particular neurons in thebrain's centers are not inherently markedfor a prescribed function or meaning. Itis adaptive that no program exists in thenervous system as a hardwired entity andthat neural maps are selected by the brainin the course of development.8

8 Reeke and Edelman caution against assumingtoo much similarity between their biological workon neural networks and the connectionist models.They state that connectionist models "'look side-ways to biology' because they take their inspirationand much oftheir terminology from the neural net-works in living organisms, but they are not modelneural networks (nor are they intended to be). Phys-icists, in their search for simplicity, are not preparedto deal with systems whose fundamental aspect liesin variability rather than regularity. In the attemptto find regularity in biological systems, many fea-tures have been introduced into their simulation inconnectionist systems that are quite unbiological.These include the notion ofmemory as a replica ortransformation of"information" given in the world(human memories are highly context- and affect-sensitive and to some extent nonveridical); the con-ception of memory retrieval as the relaxation of anetwork to a stable state (a brain is continuallyexposed to changing input patterns and has no op-portunity to freeze them while waiting for the ap-proach to equilibrium); the idea of energy mini-mization through simulated annealing (a braindecides actions more quickly than known annealingprocedures could attain in model networks oper-ating at the speeds of real neurons); the notion ofbidirectional and symmetric single connections(synaptic connections in the brain are monodirec-tional); and the idea that learning can proceed byclamping the output ofthe system to a desired valuewhile synaptic weights are adjusted according tosome rule (the motor output of a brain can in gen-eral not be imposed externally). Yet each of theseelements is present in one or another of the con-nectionist models" (Reeke and Edelman 1988, pp.152-153).The connectionist work Reeke and Edelman are

evidently referring to here is by David Rumelhart

Chomsky's point in his 1959 reviewthat no "independent neuro-physiologi-cal evidence" is available (27) might havesounded convincing 30 years ago. How-ever, in Knowledge ofLanguage, he con-tinues to assert that "so little is nowknown about the relevant aspects of thebrain" (1986, p. 39). At some moment,the point must simply be conceded thatneurophysiological advances have beenmade and that these have shown that thebrain does not work in an algorithmicmode (Edelman, 1987, p. 44). The powerofessentializing humanism is running outof steam, and the search for those genet-ically-encoded, hardwired, essential ab-solutes ofhumanness must eventually beabandonned.And what is to fear in that? Chomsky

reminds us most recently in Languageand Problems ofKnowledge:

Quite typically, intellectuals have been ideologicaland social managers, serving power or seeking toassume power themselves by taking control ofpop-ular movements of which they declare themselvesto be the leaders. For people committed to controland manipulation it is quite useful to believe thathuman beings have no intrinsic moral and intel-lectual nature, that they are simply objects to beshaped by state and private managers and ideo-logues-who, of course, perceive what is good andright. Concern for intrinsic human nature posesmoral barriers in the way ofmanipulation and con-trol .... In accordance with these conceptions, hu-man rights are rooted in human nature, and weviolate fundamental human rights when people areforced to be slaves, wage slaves, servants ofexternalpower, subjected to systems of authority and dom-ination, manipulated and controlled "for their own

and James McClelland and thePDP Research Groupwhere, in contrast to older work on AI, computersimulations of certain language learning tasks donot rely on the prior existence of rules and para-digms. As opposed to the old approach, where gen-eralizations are conceived of in terms of rule use,the new approach conceives of generalizations interms of cue acquisition.

Pinker and Prince (1988) critiqued the Rumel-hart and McClelland simulation of the acquisitionof the past tense in English. See, however,MacWhinney and Leinbach (1989) for analysis ofthe acquisition of German gender in a Rumelhartand McClelland framework.Also see Rosenfield (1988) for a nice introduction

into Edelman's work as well as a thorough critiqueof the idea of memory as "stored" images of rep-lication of experience.

56 JULIE ANDRESEN

good." (1988, pp. 165-166; see also Chomsky,1971)9

I thoroughly appreciate the moral highroad Chomsky takes here. I, too, amagainst slavery, but it is notable that ar-guments against behaviorist and othernon-essentialist accounts of human be-havior(s) have been obliged to shift froman epistemological grounding to a moralone. Non-essentialist accounts-e.g., theaccount of verbal behavior offered bySkinner-seem to force "a retreat froman epistemological position to a moralone, transforming strictures from whatcan not be known [e.g., neurophysiolog-ical brain structure] to what should notbe known [e.g., that verbal, and otherhuman, interactions are at some level andperhaps fundamentally manipulative]"(Smith, 1957, p. 2). It seems that, in themost simplistic terms, an anti-behavior-ist stance has the moral mandate to wardoff the possibility that if a behavioristaccount should happen to correspond tosome aspect ofreality, then knowledge ofit in the wronghands would be disastrous.Time and space limitations prevent mefrom fully commenting on this dimen-sion of the Skinner/Chomsky juxtapo-sition. I will only point out that nothingchanges as a result of differing accountsofhuman language, and that we have hadvicious slavery, evil bloodshed and po-litical tyrannies of various malevolentsorts without any theory of language (orhuman nature) whatsoever.

iv) The Discipline ofLinguisticsas It Conceived ofItselfThrough Its Textual Tradition

I return to Chomsky's review of Skin-ner. One of the more salient features ofVerbal Behavior is Skinner's innovativeterminology, and one ofthe more salientfeatures of Chomsky's review was hisconstant effort to wrench that terminol-ogy back into the traditional definitions.

9 1 wish to thank Kary Smout for having pointedout the relevance ofthis passage in Chomsky (1988).Chomsky (1971) might also be invoked for a dis-

cussion of the "dangers" and/or "vacuity" of be-haviorism.

At various instances, Chomsky com-ments on Skinner's use of terms: "It ap-pears that the word 'control' here is mere-ly a misleading paraphrase for thetraditional 'denote' or 'refer' (1959, p.33); "We are no doubt to interpret theterms 'strength' and 'probability' in thiscontext as paraphrases of more familiarlocutions such as 'justified belief or 'war-ranted assertability,' or something of thesort" (1959, p. 35); "The phrase 'X isreinforced by Y (stimulus, state ofaffairs,event, etc.)' is being used as a cover termfor 'X wants Y,' 'X likes Y,' 'X wishesthat Y were the case,' etc." (1959, p. 38).Chomsky's dislike of Skinner's termi-nology is thorough-going. He does notlike mand or tact or echoic operant, andhe certainly does not like anything resem-bling 'textual behavior' or 'intraverbaloperant'.'0 Writes Chomsky:A verbal response to a written stimulus (reading) iscalled 'textual behavior'. Other verbal responses toverbal stimuli are called 'intraverbal operants'. Par-adigm instances are the response four to the stim-ulus two plus two or the response Paris to the stim-ulus capital ofFrance. Simple conditioning may besufficient to account for the response four to twoplus two, but the notion of intraverbal responseloses all meaning when we find it extended to covermost of the facts of history and many of the factsof science; all word association and 'flight of ideas';all translations and paraphrase; reports of thingsseen, heard, or remembered; and, in general, largesegments of scientific, mathematical, and literarydiscourse. (1959, pp. 51-52)

"Discourse" is the very interesting wordin his passage, and the contemporaryreader familiar with Foucault will bestruck by just how similar are Skinner'snotion of the "intraverbal" and Fou-cault's understanding of "discourses,"

10 "Intraverbal behavior" is that portion of"ver-bal behavior" that Skinner claims is under the con-trol of "verbal stimuli." "Intraverbal behavior" is,thus, distinguishable from, say, a tact which wouldbe an operant "exemplified when, in the presenceof a doll, a child frequently achieves some sort ofgeneralized reinforcement by saying doll" (Skinner,1957, p. 81). The idea associated with "tact" ismaking contact with the physical world. The ideaassociated with "intraverbal" is operating withinan already verbally-defined territory of thoughts/ideas/feelings.

See Vargas (1986) for an explanation and a re-organization of "intraverbal behavior."

SKINNER AND CHOMSKY 57

these grand realms ofwords and texts andtraditions we are disposed to calling "sci-entific" and "mathematical" and "liter-ary." It is not, as Chomsky would haveit, that the notion ofintraverbal response"loses all meaning" when extended tocover these discourses; it is rather thatthese discourses can no longer make cer-tain kinds oftruth claims under Skinner'sdescription. With such a thorough dis-crediting of Skinner's terminology byChomsky, it is no wonder that Skinnerwrote some years later that Chomsky'sreview was "not really a review of mybook but of what Chomsky took, erro-neously, to be my position" (1972, p.346).11The effect ofChomsky's review was to

bar the possibility of Skinner participat-ing in the intraverbal behavior called"linguistics." Chomsky opened his re-view with terms like language andlanguage behavior and closed it with re-peated instances of the phrases present-day linguistics, grammar, theory of lan-guage, construction of a grammar, andmastery of language, thus anchoring hisreview in its own textual tradition. Andjust in case anyone missed the intraver-bal operants controlling Chomsky's dis-course, he identified them unequivocallya few years later in Aspects ofthe TheoryofSyntax. In the opening pages, after de-scribing how linguistic theory is con-cerned primarily with an ideal speaker-listener, Chomsky buttressed his positionby stating that it was consonant with thatof the "founders of modern general lin-

" I wish to thank Stephen Murray for havingkindly sent me a copy of an exchange of letters hehad with Skinner in 1977. Skinner's comments onChomsky's review ofBeyond Freedom and Dignitybear repeating. Skinner wrote in October 1977: "Ihave never actually read his long review ofBeyondFreedom and Dignity though I have read three an-swers to it none of which the New York Reviewwould publish. I have neverbeen able to understandwhy Chomsky becomes almost pathologically angrywhen writing about me but I do not see why I shouldsubmit myselfto such verbal treatment. IfI thoughtI could learn something which might lead to usefulrevisions ofmy position I would ofcourse be willingto take the punishment, but Chomsky simply doesnot understand what I am talking about and I seeno reason to listen to him."

guistics [the Neogrammarians?]," anddeclared firmly that "no cogent reasonfor modifying it has been offered" (1965,pp. 3-4). 12 Later in the chapter, and aftera passing tip ofthe hat to Saussure's lan-gue/parole dichotomy, he harkened backnot only to Humboldt but also to thegrammaires ge'nerales and accepted asgiven the problems of "linguistic theory"as traditionally posed. The advanceChomsky offered in Aspects was suppos-edly the ability to solve those problems(1965, p. 8). 13

Again, Chomsky's pursuit of a pathforged by "traditional linguistic theory"no doubt sounded less problematic in1965 than it does today. The last twentyyears ofscholarship have seen the adventand the expansion of linguistic histori-ography, and that practice has made itimpossible to point uncritically to someunitary, even monolithic, tradition thatqualifies as "linguistics" or to accept un-critically the problems posed by that tra-dition. The last twenty years of linguistichistoriography has permitted the reas-sessment and reintegration of numerous"lost" or "forgotten" or otherwise un-dervalued and disparaged ideas, ap-proaches, and/or insights to the study oflanguage that, for one reason or another,were either stricken from the historicalrecord or disbarred from it from the be-ginning. I am not saying that I wish lin-guistics to suddenly rally around Skinnerand adopt him in toto or uncritically. Iam saying that his approach is interest-ing, even salutary, and that writing Skin-ner into record changes the history ofwhat

12 See Andresen (1990, p. 184) for a reinterpre-tation of the "scientific advances" achieved by theNeogrammarians.

13 See Volosinov (1973 [1929]) for a brilliant cri-tique of"Two Trends ofThought in Philosophy ofLanguage," the two trends being "abstract objec-tivism" whose most striking expression is Saussure,and "individualistic subjectivism" whose founda-tions were laid by Wilhelm von Humboldt. ForVolosinov, "abstract objectivism" is "listener-ori-ented" while "individualistic subjectivism" is"speaker-oriented." However, Volosinov's solu-tion to the problems raised in the two opposingaccounts was not to simply put the listener andspeaker in the same skin, as Chomsky did, but toput them in dialogue.

58 JULIE ANDRESEN

we think our discipline to be and therebyreconfigures the disciplinary bounda-ries-which is, after all, the purpose ofhistoriography.The Skinner/Chomsky episode has a

fascinating, curious and perhaps evenunique place in the historical record oflinguistic activity. The principals are stillalive, although Skinner is reportedly veryill, and those ofus who do not rememberthe splash of Chomsky's 1959 reviewhave at least heard of it as part of theoral history of our discipline. Strangely,no reviews of Verbal Behavior appearedin psychology journals or other social sci-ence journals, and while all of Skinner'sother books were widely reviewed, thepsychologists were evidently just not in-terested in Skinner's work on language.As far as I can tell, the only two reviewsever published of Verbal Behavior werethe one by Chomsky in Language andthe one by 0. K. Tikhomirov, also in1959, in the journal Word. Since Tik-homirov's review was also generally neg-ative, Skinner's oblivion in languagestudies seemed secured.Given that Tikhomirov was a Soviet

behaviorist in the Department of Psy-chology at the University ofMoscow, hiscriticism of Skinner was, not surprising-ly, entirely different from Chomsky's.Tikhomirov followed Pavlov who distin-guished two systems of signals, the firstbeing the signal system of reality whichman shares with animals, while the sec-ond is attuned to Marxist/social under-standing:The very process offormation of speech activity inman does not manifest the regularities of behaviorformation of animals. Man as a social being ac-quires socially determined patterns ofbehavior, in-cluding speech behavior. This process of acquisi-tion signifies a new type ofdevelopment in humanbehavior and is due precisely to the social existenceof man. Furthermore, the acquisition of languagebecomes the subject of special training (in schools,for example) and consequently cannot be reducedto individual activity. (1959, p. 366)

We see here, then, a fifth -and final-reason why Skinner's approach did not"take" in the late 1950's. While, on theone hand, Skinner failed to "properly"acknowledge autonomous human agencyand an autonomous human language

ability demanded by Chomsky, he alsofailed to "properly" foreground theMarxist/social dimension of human be-havior demanded by Tikhomirov.

Nevertheless, many of Skinner's for-mulations do not seem radically at oddswith the important study, Marxism andthe Philosophy of Language, written byanother Soviet, V. N. Volosinov, whoformulated a "behavioral ideology" (1973[1929], p. 91ff). Indulging in the kind ofpost hoc speculation that is the peculiarspecialty of the historiographer, I willhazard to say that Skinner would havefound many points of satisfaction in Vo-losinov's account of language. If we putSkinner and Volosinov in dialogue, it ap-pears that:

a) Skinner probably would have"agreed" with Volosinov's formulationthat: "individual consciousness is not thearchitect of the ideological superstruc-ture, but only a tenant lodging in the so-cial edifice of ideological signs" (1973[1929], p. 13). It is certainly in accordwith Skinner's notion of the individualas "locality" rather than "prime actor."

b) Skinner would have been sympa-thetic to Volosinov's discussion of thattroublesome concept "meaning." Let uscompare Volosinov's statement:

... there is no reason for saying that meaning be-longs to a word as such. In essence, meaning belongsto a word in its position between speakers; that is,meaning is realized only in the process of active,responsive understanding. Meaning does not residein the word or in the soul of the speaker or in thesoul of the listener. Meaning is the effect of inter-action between speaker and listenerproduced via thematerial ofa particular sound complex. It is like anelectric spark that occurs only when two differentterminals are hooked together. (1973 [1929], pp.102-103)

to Skinner's:

It is usually asserted that we can see meaning orpurpose in behavior and should not omit it fromour account. But meaning is not a property of be-havior as such but of the conditions under whichbehavior occurs. Technically, meanings are to befound among the independent variables in a func-tional account, rather than as properties of a de-pendent variable. When someones says that he cansee the meaning of a response, he means that hecan infer some of the variables of which the re-sponse is usually a function. The issue is particu-larly important in the field ofverbal behavior where

SKINNER AND CHOMSKY 59

the concept of meaning enjoys unusual prestige.(1957, pp. 13-14)

That is, for neither Skinner nor Volosi-nov could "meanings" be properties of"things." "Meanings" could only be "by-products" of "speech episodes" or "di-alogic events," which brings me to thepoint that:

c) Volosinov valorized the notion ofthe "dialogic" over and against the"monologic." The actual reality of lan-guage-speech is not the abstract system oflinguisticforms," writes Volosinov, "andnot the isolated monologic utterance, andnot the psychophysiological act ofits im-plementation, but the social event of ver-bal interaction implemented in an utter-ance or utterances" (1973 [1929], p. 94).Verbal interaction, for Volosinov, is thebasic reality of language, which includesdialogue in the narrow sense ofthe word,but also dialogue as verbal communica-tion ofany type whatsoever. A book, forinstance, is for Volosinov a "verbal per-formance in print" (1973 [1929], p. 95).

Volosinov's idea that language is a"continuous process of becoming" andhis related idea that individuals "do notreceive a ready-made language at all,rather, they enter upon the stream ofver-bal communication" (1973 [1929], p. 81)would both certainly have received ap-proval from Skinner. They have alsofound a recent voice in Paul Hopper andhis work on "emergent grammar" (1988).And, finally, d) in denying that lan-

guage is handed down as a ready-madeproduct, Volosinov avoids the reificationofthe system of language and attacks thehypostasizing tendencies of certain typesof language studies (1973 [1929], p. 81).Skinner could only shake his head in dis-may over the failure-called "hyposta-sis" -to recognize that the conditions re-sponsible for an expression such as It isfour o'clock (called a "response") may notshare anything in common with the con-ditions responsible for the response onthe part ofthe describing scientist (1957,p. 18). Perhaps Skinner is most useful tocurrent linguists in having understoodthat the concept of"language" itselfneedsquotation marks.

REFERENCES

Andresen, Julie. (1988). Review ofPaul FriedrichTheLanguage Parallax. LinguisticRelativism andPoetic Indeterminacy (1986). Language in Soci-ety, 17 (4), 600-604.

Andresen, Julie. (1990). Linguistics in America1769-1924: A critical history. London: Rout-ledge.

Austin, John Langham. (1911-1960). (1975[1962]).How to do things with words, 2nd ed.Cambridge, Mass.: Harvard University Press.

Bullock, T. H. (1967). "Signals and neuronal cod-ing." The neurosciences: A study program ed. byGardner Q. Quarton, Theodore Melnechuck,Francis 0. Schmitt and the associate and staff ofthe Neurosciences Research Program, pp. 347-352. New York: Rockefeller University Press.

Chase, Philip, & Parrott, Linda, Eds. (1986). Psy-chological aspects oflanguage. The West VirginiaLectures. Springfield, Ill.: Charles Thomas.

Chomsky, Noam. (1959). Review of Skinner(1957). Language 35, 26-58. (Repr. with intro-duction in Readings in thepsychology oflanguageed. by Leon A. Jakobovits & Murray Miron (pp.142-171). Englewood Cliffs, N.J.: Prentice-Hall,1967.)

Chomsky, Noam. (1965). Aspects ofthe theory ofsyntax. Cambridge, Mass.: M.I.T. Press.

Chomsky, Noam. (1971). The case against B. F.Skinner. Review ofBeyond Freedom and Dignity(New York: A. A. Knopf, 1971), by B. F. Skinner.The New York Review ofBooks, 17, 18-24.

Chomsky, Noam. (1980). Rules and representa-tions. New York: Columbia University Press.

Chomsky,Noam. (1986). Knowledgeoflanguage:Its nature, origin and use. New York-Westport-London: Praeger.

Chomsky,Noam. (1988). Languageandproblemsofknowledge. TheManagua lectures. Cambridge,Mass. & London: M.I.T. Press.

Dreyfus, Hubert, & Dreyfus, Stuart. (1988). Mak-ing a mind versus modeling the brain: Artificialintelligence back at a branchpoint. Daedalus, 117(1), 15-43.

Edelman, Gerald. (1987). Neural Darwinism: Thetheory of neuronal group selection. New York:Basic Books.

Gleick, James. (1987). Chaos: Making a new sci-ence. Harmondsworth, Middlesex: Penguin.

Hopper, Paul. (1988). Emergentgrammar and thea priori grammar postulate. In Deborah Tannen(Ed.), Linguistics in context: Connecting obser-vation and understanding, lecturesfrom the 1985LSA/TESOL andNEH Institutes (pp. 117-134).Norwood, N.J.: Ablex.

Koemer, E. F. K. (1983). "The Chomskyan Re-volution" and its historiography. Language andCommunication, 3, 147-169.

MacWhinney, Brian, & Leinbach, Jared. (1989).Language learning: Cues or rules? Journal ofMemory and Language, 28, 255-277.

Marr, David. (1982). Vision. San Francisco: W.H. Freeman.

Newmeyer, Frederick J., & Emonds, Joseph. (1971).

60 JULIE ANDRESEN

The linguist in American society. Papersfrom theSeventh Meetingofthe ChicagoLinguistic Society(pp. 285-303). Chicago: Chicago Linguistic So-ciety.

Papert, Seymour. (1988). One Al or many? Dae-dalus, 117 (1), 1-14.

Pinker, S., & Prince, A. (1988). On language andconnectionism: Analysis of a parallel distributedprocessing model of language acquisition. Cog-nition, 28, 73-193.

Reeke, George, & Edelman, Gerald. (1988). Realbrains and artificial intelligence. Daedalus, 117(1), 143-173.

Rosenfield, Israel. (1988). The invention ofmem-ory. A new view of the brain. New York: BasicBooks.

Rumelhart, David, & McClelland, James, & thePDP Research Group. (1986). Parallel distrib-uted processing: Explorations in the microstruc-turesofcognition. Cambridge, Mass.: M.I.T. Press.

Skinner,B.F. (1938). Behavioroforganisms. NewYork-London: Appleton-Century.

Skinner, B. F. (1957). Verbalbehavior. NewYork:Appleton-Century-Crofts.

Skinner, B.F. (1966). The phylogeny and ontog-eny of behavior. Science, 153, 1205-1213.

Skinner, B. F. (1972). A lecture on 'having' a poem.Cumulative record:A selection ofpapers, 3rd ed.,pp. 345-355. New York: Appleton-Century-Crofts.

Skinner, B. F. (1981). Selection by consequences.Science, 213, 501-504.

Smith, Barbara Herrnstein. (1957). A new anat-omy of language, Unpublished manuscript re-view of Skinner (1957).

Tikhomirov, 0. K. (1959). Review of Skinner(1957). Word, 15, 362-367.

Vargas, E. A. (1986). Intraverbal behavior. InChase & Parrott (pp. 128-15 1).

Volosinov, Valentin N. (1973 [1929]). Marxismand the philosophy of language, translated byLadislav Matejka and I. R. Titunik. Cambridge,London: Harvard University Press.

This is a reprint of Julie T. Andresen, "Skinner and Chomsky Thirty Years Later," published inHistoriographia Linguistica XVII: 1/2 (Amsterdam & Philadelphia: John Benjamins Publ. Co., 1990),pp. 145-166.