Re: Chomsky vs. Skinner

From: Stevan Harnad (harnad@cogsci.soton.ac.uk)
Date: Sat May 25 1996 - 19:57:05 BST


> From: "Bollons Nicholas" <NSB195@psy.soton.ac.uk>
> Date: Thu, 11 Apr 1996 13:43:54 GMT
>
> HOW DOES SKINNER AND CHOMSKY'S EXPLANATION FOR THE ACQUISITION OF
> GRAMMAR DIFFER ?
>
> Skinner's explanation of language was that any
> acquisition was due to a learning process involving the shaping of
> grammar into a correct form by the re-enforcement of other stimulus,
> correct grammar is positively re-enforced and will be used in the
> future, and incorrect grammar is negatively re-enforced and will be
> not be used again. Chomsky differed in his view that human grammar
> acquisition is an innate biological ability that all humans possess,
> and viewed some form of `generative grammar' which he felt could
> explain the rapid acquisition and creative nature of grammar and
> language.

Chomsky's Universal Grammar (UG) explains our grammatical capacity.
It turns out that UG cannot have been learnt by Skinnerian trial and
error because of the "poverty of the stimulus." The child never hears or
produces all the correct and incorrect sentences that would need to be
sampled in order to learn it all by trial and error and reinforcement.

> Skinner was a Behavioural Psychologist and was unconcerned with any
> `underlying mental processes' that may have been occurring during
> learning and denounced these `un-observables' as fictitious. Instead
> he was concerned with the observable materialistic nature of
> behaviour. He felt that there was no `underlying' meaning to words
> and that verbal behaviour was due to the conditioning that occurs
> between the words and the reinforcement properties of a stimulus.
> This does not just apply to physical reinforcement stimulation: `that
> if you ask for "a glass of water", you get one', but also social
> reinforcement in the form of praise and encouragement i.e. `well
> done'. This reinforcement is an important concept, and past
> experiences of verbal behaviour are important in determining whether
> they will be used again. Skinner used the phrase `Verbal Operant
> Conditioning' where a verbal response that occurs in a given
> situation and is followed by a reinforcer becomes more likely to
> occur again in the same situation. Skinner identified five separate
> classes of `Verbal Operant' : Mands, Tacts, Ethoics, Textuals and
> IntraVerbals ( cited in Psychology of Language : Paivio & Begg 1981).
> He also explained grammatical acquisition using an `Autoclitic', but
> seeing as the `Autoclitic' uses all five of the `Verbal Operants', a
> brief explanation of their characteristics had better be given first.
>
> The Mand is based on the speaker conveying a command or request for
> something which is met with it's production. A child knows from it's
> past history that if it says "milk" it will most probably get milk
> (milk becomes the reinforcer). The Mand is rooted in the idea of
> `com-Mand' and `de-Mand'. Tacts on the other hand, are based on
> child's reference to non-verbal objects and the use of `naming'.
> Reinforcement occurs through the praise for correct naming i.e.
> child says "cat" and mother says "good girl, that is a cat". The
> Echoic Operant is the imitation of speech and can be reinforced by a
> variety of means including "repeat after me". Textual is the reading
> part of the `Verbal Operants'. The last, but by no means the least is
> the Intraverbal operant. This is a form of `word association' where a
> certain word will control which word is to proceed after it, that
> only certain words can go after the word that has just been said (a
> sort of "table, chair" game but involving all types of words).
> According to Skinner this process is going on all the time whenever
> we use a word, only a certain set of words can go after it.
>
> So we arrive at the `Autoclitic'. This is a form of commentary made
> upon one of the `Verbal Operants' described above. Correct grammar
> uses the correct order of `verbal operant' and Autoclitic comment.
> But Autoclitic comments also take on the form of having verbal
> operant characteristics. An example (taken from Psychology of
> Language : Paivio & Begg 1981) is probably needed : `the primary
> [verbal] operant is the tact `John is in Montreal'. If the speaker
> says " I hear that John is in Montreal" we have an example of
> autoclitic words in which `I hear that' is a comment on a primary
> [verbal] operant that is presumably an echoic (that is, the speaker
> heard that someone say that John is in Montreal'). Still confused ?
> `John is in Montreal' is the primary `part' of the sentence
> involving the words being produced individually using the verbal
> operants. Then the sentence is commented upon by the `autoclitic'
> which itself has the characteristic of being echoic. If the
> sentence was "I read in the paper that John is in Montreal" then the
> autoclitic comment would have a `textual' (reading) characteristic.
> Correct Verbal Operant and Autoclitic comment (correct grammar) will
> be positively reinforced by praise e.t.c and said again, whereas
> incorrect grammar will be negatively reinforced ; `no that's wrong'
> and not be said again (such as a child who says a gibberish sentence
> like `Montreal read in paper that is John I the') . The complexities
> of studying such a model, in dissecting each sentence into it's
> appropriate verbal operant and autoclitic parts, abound.

Good kid-sib summary...

> So what evidence have Skinner and Behaviourists produced to validate
> their claims? Guess et al 1968 (cited in Psychology of Language :
> Paivio & Begg 1981) described how they taught a mentally retarded
> girl to make correct grammatical utterances using positive
> reinforcement of praise and food. Studies on adults giving
> reinforcement for certain nouns and plurals, found an increase in an
> occurrence of correct responses, if `praise' was issued (see Holtz &
> Azrin Conditioning Verbal Behaviour 1966). In general, children are
> considered to acquire correct grammar through reinforcement of their
> verbal teachers (particularly their mothers). Yet there is little
> evidence to support such a claim. Not only has it been found that
> there is no relationship between child correct grammar and parental
> positive reinforcement it seems that parents are only interested in
> the correctness of a child's meaning (see Slobin 1975). Also some
> adult subjects are un- aware of the relationship between correct and
> incorrect responses and the reinforcement that they receive because
> of their grammar ( see Konecni & Slamecka 1972). Reinforcement does
> not seem to occur in the right context but it also seems not to be
> noticed. Many (including the linguist Noam Chomsky) have advocated
> that a behavioural stimulus - response system involving
> reinforcement and a `finite grammar' model, cannot explain the
> rapid, creative and complex nature of language. Chomsky `stepped
> into the limelight' in the debate on language and grammar
> acquisition in his Review Of Verbal Behaviour by B.F Skinner
> (Chomsky 1959). It is to him that attention will now be turned.
>
> Chomsky argued that there was no way that a child can obtain a
> language from only the `primary linguistic data'(Psycholinguistics
> 2nd Edition : Slobin 1969) that the child receives from it's
> teachers and environment. He prescribed that an infant enters this
> world with a predisposition to learn a language fluently, and this
> predisposition is encased in our biological make-up, innate to all
> humans. (A sort of `Language Acquisition Device' or `L.A.D' as it has
> now become known). One of the manifestations of this was in the form
> of a `generative grammar', that had the ability to `generate' and
> create all the words in a linguistic grammar that he viewed
> Skinner's `finite grammar' (though capable of producing) was far too
> limited in it's application.

Remember that Chomsky did not define or assume any of this: He concluded
it from his data and the theory he needed to explain it. It turns out
that the rules you need to "know" to be able to do what the 4-year-old
can do with word sequences cannot be learned, because the stimuli
are not there. Therefore they must be inborn.

> Chomsky defined this `generative grammar' as : `finite set of rules
> operating on a finite vocabulary to generate an infinite number of
> acceptable grammatical sentences and no un-acceptable ones' (quoted
> from Psychology of Language : Paivio & Begg 1981). So, from a small
> number words, using `some rules', we can create a vast (infinite)
> number of words, you can in fact boil down all the words in the
> English Language to around 500 or so. What did the form of this
> generative grammar take the shape of , and what are these `some
> rules' that are mentioned above? Chomsky explained the acquisition
> of grammar using a process of transformation or `Transformational
> Grammar' which is perhaps `the physical form', and the rules that
> are `generative grammar'. As may be apparent `transformational
> grammar' involves the changing of sentences into other states. They
> are metamorphosed into their `deep' and `surface' structures by a
> set of rules, or phrase structures. These two `deep' and `surface'
> structures are then interpreted according to their phonological and
> semantic meaning. The theory is considerably more complex than can
> be explained here, and has been revised many times by Chomsky since
> it's first publication in 1957 (see Chomsky 1965). The result of
> this, is that once a child can master these rules and
> transformations, it has the ability to create and expand on his/her
> grammar by using these rules to create new sentences that it has not
> heard before (which `finite grammar cannot as the child needs to use
> or hear the word and find out it's reinforcement properties). A
> great deal of creativity occurs in child grammatical utterances.
> Chomsky viewed this creativity as a very important aspect.
>
> So what evidence is there to support Chomsky's view of
> `transformational generative grammar' and that this is an innate
> biological species specific trait ? All languages of the world share
> similar characteristics of using nouns, verbs, pronouns, though not
> necessarily in a similar order. Grammar and complex language usage
> seem also to be a `uniquely human capability' as no other species on
> the planet seem to posses such proficiency as humans, though there
> have been some successes in teaching `sign language' to Chimpanzees,
> it is viewed that any ability that they grasp, cannot be seen as
> Homologous, similar in structure, to the complex human abilities of
> language (see Modularity, Domain Specificity and the Development of
> Language : E. Bates).

And if you don't believe that, try discussing it (or anything) with a
chimp...

> One theoretical concept, so abstract, so
> explicit, and usually overlooked, is proposed by S. Pinker & P. Bloom
> in their article Natural Language And Natural Selection (Behavioral
> and Brain Sciences 1990). The fact that all human languages use
> symbol manipulation to make references to something else that they
> (the symbols) are not physically related to, i.e. some form of
> `arbitrariness', is an important aspect . The sign is only arbitrary
> to the thing it comes to represent. For example, the sign for `Genus
> Canus' can be dog, chien, skili or even ##stlg, it does not matter as
> long as the word, or sign used, comes to represent the object/concept
> and is in no way physically related to it. ( Onomatopoeia will not be
> discussed). But say that every time you wanted to talk about `Genus
> Canus' you had to produce one, or draw an exact copy, this is not
> arbitrary or particularly practical. Pinker & Bloom defined this as
> some form of `universal grammar' that is an innate ability in all of
> us that uses this `arbitrary symbol manipulation' adapted to it's
> present complex form by the process of Natural Selection. This human
> symbol manipulation similarity is an important aspect (Watch out do
> not confuse your grammars : `universal grammar' (arbitrary symbols )
> is different from generative grammar ( rules used to manipulate these
> arbitrary symbols).

Not quite! Universal Grammar (UG) is the 1990's version of what used to be
called Generative Grammar. UG is one particular symbol system; there are
many others. Not all of them are natural languages. Arithmetic is an
example; formal logic is another; the C++ programming language yet
another. They all share the property, though, that their symbols are
arbitrary in their shape; their shape does not resemble what they stand
for. And the symbols are just manipulated on the basis of their shapes,
not on the basis of what they stand for.

UG is a symbol system that explains our grammatical capacity.
Originally, Chomsky thought that natural language grammar was
independent of meaning, i.e., the rules of grammar are just based on the
shapes of the sentences, not what the words or sentences mean. This was
the origin of "modularity," the idea that certain cognitive capacities
might be independent of one another, and could be explained
independently. In modern UG it is no longer entirely clear whether
syntax is entirely independent of semantics.

> Other evidence to support Chomskian claims, can be found in the
> study of language deficient patients (aphasia). It has long been
> known that damage to the left hemisphere near the motor cortex
> (Broca's Area) causes a loss of speech, especially to grammatical
> elements. One patient could not access the word `would' (grammatical
> word), but could access the word `wood' (content word) (see Marin Et
> Al 1976 cited in Psycholinguistics D. I Slobin). Is this
> localisation of a grammatical area in the left hemisphere ? Further
> evidence (using P.E.T) has indicated that a conclusion of this
> nature is much too naive.

That's evidence for the brain basis of language, but not particularly
for the brain basis of UG. Even if Skinner had been right, and grammar
was learnt, it would still have had a brain basis eventually, and brain
damage might have affected it.

> There is a volume of evidence to support Chomsky's claims of an
> innate generative grammar, though it is difficult (and unscientific)
> to introspect whether a transformational process occurs during
> grammar acquisition, this would seem considerably more likely than a
> `finite' system. These days `generative grammars' have become a
> widely accepted theory in linguistics and cognitive science, for
> they manage to answer more of the questions on the rapid and
> creative nature of grammar acquisition than a Skinnerian, or even a
> Neo Behaviourist, model can see fit. (Of course behaviourist's do
> not feel this way and defend their claims adamantly - see What Are
> The Scope Of Radical Behaviourist Theory : Questions To B.F Skinner
> By S. Harnad - question F). Recently, contemporary debate has been
> focused upon the nature of this Language Acquisition Device, which
> Chomsky proclaims. Just how much innate language ability does a
> child have when he/she enters this world ? (see Modularity, Domain
> Specificity And The Development Of Language 1994 - E. Bates). What
> are the semantic (meaning) components of transformational, and most
> grammar systems used in language? (This is of particular importance
> to Psychologists). The detailed and vast nature of language, the
> difficulty in collecting empirical data during child language
> acquisition and the fact that language touches on so many areas of
> Psychology, make validation of Chomsky and Skinner's theories, and
> any language theories, particularly difficult. Though it would seem
> more favourable, from the evidence, to accept `some parts' of
> Chomsky's theory.

Good job, Nick!



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:24:16 GMT