Psychology Naom Chomsky Review of Skinner s Verbal Behaviour Notes

Transcription

1 Section 1 () Many philosophers and linguists had hoped that study of language could be incorporated into behaviourist psychology as a way of opening up difficult areas (such as meaning) to fruitful investigation. Skinner s Verbal Behaviour was the first large-scale attempt to do this. Behaviourism gives a functional analysis of verbal behaviour in terms of variable, control, specification, interaction, response, stimulus, reinforcement, deprivation and suchlike. The goal of Skinner s book is to provide predictions of ways to control verbal behaviour via the physical environment of the speaker. Skinner thinks that we can extend experiments on the behaviour of animals to humans with very little modification (a view not widely shared!). Other behaviourists don t share Skinner s confidence in the applicability of animal studies to complex human behaviour. What s remarkable about Skinner? Not functional analysis and input / output observable per se, but the limitations on the approach to study and the simple functions claimed. Skinner ignores internal organisation (innate, genetically-determined motivation and experience). This is a necessary approach if we don t have neurophysiological information, but an estimate of the importance of internal versus external investigation is critical. We end up stuck with explaining output in terms of input: nothing controversial about Skinner s position here. Skinner denies the importance of the contribution of the organism given that this is complex, successful prediction of behaviour will then depend on very indirect and detailed study of behaviour. Skinner s thesis is that external stimulus plus reinforcement is what explains complex verbal behaviour. He claims that the speaker s contribution is trivial and that the prediction of verbal behaviour only requires the specification of external factors, as in the case of lower organisms. This isn t justified. Reinforcement only applies to complex human behaviour in the most superficial way. Skinner omits the fundamental factors that are amenable to scientific study. The magnitude of Skinner s failure is a measure of the importance of the factors he omits from consideration and of the depth of our ignorance. Chomsky s approach is to review Skinner s examples, in which lie the force of his argument, and also the concepts in which functional analysis is presented. Skinner s basic concepts are stimulus, response and reinforcement. Chomsky s positive contribution in this review will be to consider how further linguistic work might clarify some of the issues raised by Skinner. Section 2 () Skinner divides animal responses into two categories: 1. Respondents: reflexes. 2. Operants (where response is emitted without obvious stimulus). Operants are Skinner s primary area of interest. Description of the Skinner Box. Some technical terms follow. Theotodman@lineone.net Page 1 of 6

2 Reward increases the strength of operant behaviour (reward is a reinforcer, given in a reinforcing event). The strength of an operant is defined as the rate of response during extinction (where extinction is the time between the last reinforcement and the return to the pre-conditioning rate of response). Conditioning: the reinforcer is only given if the agent acts on the particular stimulus. Stimulus discrimination leads to a discriminated operant. Eg. press the bar when the light flashes. The flash is the occasion where the operant is emitted. Can also have response differentiation, where the response has to be of a particular character ( press the bar for this long ). Chomsky thinks that the distinction between the eliciting stimulus, discriminating stimulus and originating forces has never been made clear. There s a lack of experimental control for private internal events. Can quickly change the character of the animal s response by tweaking the reinforcing event (what the animal must do to be rewarded) this can result in complex behaviour. Secondary reinforcers are associated with already reinforcing stimuli (money, as associated with food). While it s possible that chimps are willing to earn money by reinforcement with food, it s plausible that humans are motivated by money in this way. Any view can claim support given ambiguities. Drive is defined operationally in terms of hours of deprivation (of the reinforcer). Chomsky notes that the difficulties of generalisation are shown by the differences between rats and fish in the solution of elementary problems. There are definitional questions outstanding. Is the stimulus only something to which the agent in fact reacts, or any (actual) physical event to which it s capable of reacting? Is a response only counted if it is connected to the stimulus in lawful ways? We have a dilemma: 1. Broad definitions of stimulus and response lead to no demonstration of lawful behaviour (ill defined attention, volition and caprice). 2. Narrower definitions make behaviour lawful by definition, but this is of limited applicability because most of what the animal does will not be considered to be behaviour. So, either we can t show behaviour to be lawful (maybe this isn t too bad for a developing science) or we have to restrict our attention to good experiments in which it is lawful. However, Skinner is inconsistent. Analogic guesses in terms of metaphors extending his technical vocabulary. This gives the illusion of a rigorous broadbased scientific theory, but real life isn t like the laboratory. Effectively use homonyms. To justify this accusation, Chomsky must show that, taken literally, Skinner s technical vocabulary has no application to linguistic behaviour, and that the metaphorical reading leaves us worse off than the traditional approach. Stimuli are mixed up with the resulting verbal behaviour and are consequently not capable of manipulation. Skinner s laws of behaviour are untestable or of limited scope, and his notion of lawful relation is arbitrary and obscure. Skinner is taken to be good at experimentation, but bad at theory. Theotodman@lineone.net Page 2 of 6

3 Section 3 () Stimulus and responses have to be lawfully related if these terms are to be used. This is not generally so with human behaviour. Skinner is inconsistent in accepting tight law-like links required from stimulus to response, but simply ignores this constraint in his discussion of language. Chomsky refers to the variety of responses that a picture might elicit the response Skinner expects Dutch when a speaker is confronted by a Dutch Master is by no means the only one. It s a simple but empty device to say that the picture elicits responses to its subtle properties, because any number of them are on offer. We ve lost all objectivity of the word stimulus it has been driven back into the organism and is no longer part of the environment. We can only identify the stimulus when we hear the response. Talk of stimulus control disguises a retreat to mentalistic psychology. If we don t know the stimulus until we hear the response, we can t predict verbal behaviour. In a footnote, Chomsky gives the sarcastic pencil sketch approved of by Barry Smith in order to elicit the response pencil. So, Skinner s system doesn t permit control of verbal behaviour. Proper names aren t responses to stimuli by particular things. For instance, if we ve never encountered the thing (Moscow). Also, the probability of uttering someone s full name isn t increased by their presence (would we ever do this?). Skinner s logic would mean that one s own name wasn t a proper name. Control is used where we d normally use denote or refer. It is wrong to say that the relation of reference is the probability that a speaker will respond to stimulus, at least if we use these technical terms in their literal sense. Chomsky gives a selection of Skinner s examples of stimulus control. There s no relation to bar-pressing in the Skinner Box, and most of the alleged stimuli aren t even present. Skinner is satisfied with a vague and useless answer to the question what is a verbal operant. It is allegedly related to controlling variables, but Skinner gives no means of determining what these are. There are no answers either to other elementary questions such as when are two operants the same. Skinner is content to extrapolate from the laboratory without answering detailed questions. The laboratory allows the easy identification of units of behaviour & the operant is defined by reference to a particular experimental procedure, but this can t be extrapolated to real life. Skinner defines response strength as the rate of emission during extinction. How can this apply to speech? Is probability of emission related to frequency? Skinner actually uses factors other than probability or frequency to indicate the strength of response. Speed and energy are suggested, inter alia, but Chomsky shows how a quiet response may be more effective. An awed whisper may be better than a repeated shriek! Response strength is the basic datum in functional analysis. Skinner s use of probability is an objective cover-up for intention or belief. As an example, Chomsky cites Skinner s suggestion that confirming assertions in science are said to be by increasing its probability and strength. If these words are given their technical sense, this is ludicrous it implies crowds shouting! Theotodman@lineone.net Page 3 of 6

4 Chomsky thinks he s shown that Skinner uses laboratory or technical terms such as stimulus or response etc. as paraphrases for normal mentalistic descriptions without any increase in objectivity. As he s not using these laboratory terms in their primary sense, he s made no progress. Section 4 () Chomsky now turns to Skinner s use of the term reinforcement, which he thinks raises even more serious problems. There s no problem with Skinner s definition of reinforcement in the context of the laboratory, but it s useless in real life, because he needs to characterise the stimuli that are reinforcing otherwise than by the mere fact that they are. Reinforcement is circular as used by Skinner in the law of conditioning. Learning is taken as change in response strength, but vagueness in the term reinforcement makes the claim unverifiable. Skinner doesn t even insist that a reinforcer be an identifiable stimulus. Chomsky shows this vacuity by example. Skinner imagines automatic selfreinforcement (repetition in private). Thinking is behaviour that automatically reinforces the behaver since it affects him (Chomsky notes as a reductio that cutting ones finger is reinforcing and should therefore be an example of thinking). Self-reinforcement by production of speech, writing, playing or painting ( practice ). Reinforcement can also be by speaking, remaining silent or acting. The strength of behaviour is measured by listeners response, leading to communication letting the listener know. The agent is not always even present when he is reinforced such as when his work is appreciated by others in his absence, or after his death! And so on The term reinforcement has for Skinner lost all objective meaning. The person reinforced need emit no response and the reinforcing stimulus need not exist or impinge on the reinforced person. Skinner claims that we do what we enjoy because we find it reinforcing, or hope that others will be reinforced. Thus, Chomsky points out that the term reinforcement can have no more than a ritual function. X is reinforced by Y is a catch-all for X wants Y, etc. Reinforcement has no explanatory power and obscures the difference between the notions being paraphrased. What we want is for everyday psychological terms to be clarified or explained in terms of notions developed from careful experiment and observation. There is no conceivable interest in terminological revision in which a laboratory term is borrowed but left with all the vagueness of ordinary vocabulary. Skinner s use of conditioning (extended from its Pavlovian and operant form) is vacuous when extended to human instruction, since it adds nothing to our knowledge of such processes (and the use of conditioning in the literal sense is quite false). According to Skinner, predication facilitates the transfer of response from one term to another, but this says nothing of any significance 1. Skinner is just playing at science. 1 Chomsky gives the example of The telephone is out of order, which I ve not followed in detail. Theotodman@lineone.net Page 4 of 6

5 Sections 5-10 () Omitted from the extract Section 11 () Skinner s terms, then, when taken literally, cover no aspect of linguistic behaviour and, when taken metaphorically, obscure important distinctions. Skinner thinks it a coup to eliminate the contributions of the speaker and hearer, but this is at the cost of eliminating all significance from the descriptive system. We need to know more about verbal behaviour before attempting to answer such questions, which are hopelessly premature. We need to know what is acquired before speculating on the process of acquisition. Skinner s functional analysis question about linguistic behaviour is too difficult given current knowledge. An utterance is not a string of responses controlled by stimulation and association. The syntactic organisation of an utterance is not represented simply by the physical structure of the utterance itself. The task of contemporary linguistics is to characterise (rather than provide a complete account of) the integrative processes, imposed patterns and selective mechanisms involved in constructing an utterance. Grammar (taken to include phonology) can generate sentences as a deductive theory generates theorems. A theory of language should:- 1. Study the formal properties of grammars. 2. Determine the structure of a sentence to show how used and understood. Then, we should be able to answer the questions noted earlier about sentence construction. Chomsky divides rules into optional and obligatory. Only the latter are used in utterance generation. Optional rules are selective mechanisms 2. Semantic Generalisation 3. Syntactic investigations are a corrective to an oversimplified approach to the theory of meaning. New sentences are recognised because they are generated by internalised grammar and understood partly because we can determine the process by which the sentence is derived from the grammar. The speaker has to select a particular compatible set of optional rules. If grammatical study reveals the choices available and compatibility conditions we can investigate the factors determining the choice. The ability of humans exceeds our current understanding. The child constructs grammar based on observation and corrections. This is extremely complex and abstract, and a remarkable type of theory construction. Our theory of language acquisition must account for the fact that all children of all levels of intelligence learn language in a very short time. 2 Not clear what Chomsky means here. I think this has to do with the hypothesis testing or parameter setting that selects one grammar from another from the starting point of the innate Universal Grammar. 3 Follow this up later. Theotodman@lineone.net Page 5 of 6

6 It is difficult to accept the fair description of what s going on that the child is capable of constructing the complex sentence-generating mechanism, and an adult of using such an abstract deductive theory to determine whether a particular item can be generated from it. But, if correct, we can predict that accounts of speakers which don t include the structure of grammars are bound to fail. Grammar is inferred as a component of the behaviour of a speaker or listener only from the resulting physical acts. The fact that all children quickly acquire similar grammars of great complexity implies that humans are designed to do this. There exists a built-in information processing system to enable the construction of an adequate grammar in the time available from the available data. We cannot omit the contribution of the speaker and attribute language acquisition to generalisation without missing off everything of interest and leaving verbal behaviour a mystery. Theotodman@lineone.net Page 6 of 6